It expands ChatGPT functionality as a generative AI tool for business users. ChatGPT Enterprise is a significant step toward the commercial adoption of the new field of generative AI. It offers consistent access to GPT-4 and a variety of new and improved ...

ChatGPT for Enterprise: A Paradigm Shift in AI Adoption?

OpenAI's recent update may spark fear among many AI investors and founders who offer easily replicated services.

A minor ChatGPT update is a warning to founders: Big Tech can blow up your startup at any time

ChatGPT continues to dominate the market despite facing competition from Google's Bard and Microsoft's Bing. But surprisingly, it is not the one which brings in the most revenue!

Shock to the system! ChatGPT revenue jumps, but gets beaten by other AI apps

Pal - A ChatBot Client is an iPhone app that gets you access to the GPT-4 model for a cheaper price than the $20 ChatGPT Plus subscription.

This ChatGPT iPhone app lets you use GPT-4 for way less than ChatGPT Plus

Multi-modal mode, currently in beta, means you won't have to specify to ChatGPT to run in DALL-E 3 or web browsing mode.

ChatGPT's Latest Features Include File Analysis and Multi-Modal Support

Studycafe provides a range of industry-specific modern courses on ChatGPT, AI Tools, Business Automation, Power BI, Advance Microsoft Excel and more.

Studycafe Courses Unveils Premier ChatGPT Program: Shaping the Next Generation of Students and Professionals

Shrug Emoji It's no secret that large language model (LLM) -powered generative AI tools like OpenAI's ChatGPT — which spit out text not by way of human-esque understanding, but by way of predictive math — have a serious hallucination problem.

OpenAI Says It’s Fine If ChatGPT Occasionally Accuses Innocent People of Crimes

AI-powered chat support has made its way to WhatsApp beta, allowing for more conversational and quick responses to user queries. Meta plans to leverage AI across its platforms, including Messenger and Instagram, to enhance user experiences and generate/edit images.

WhatsApp might use AI to answer your support queries

The initial surge of excitement and apprehension surrounding ChatGPT is waning. The problem is, where does that leave the enterprise? Is this a passing trend that can safely be ignored or a ...

How knowledge graphs improve generative AI

ChatGPT and Bard lack effective defences to prevent fraudsters from unleashing a new wave of convincing scams by exploiting their AI tools.

ChatGPT and Bard risk ‘new wave of scams’, Which? warns

MobileGPT

mobilegpt

The year 2023 has undeniably marked significant advancements in the realm of Generative AI development. Just last week, I had the pleasure of attending a conference Professor Tshilidzi Marwala, an esteemed expert in the AI field. featuring Drawing from his extensive knowledge, he highlighted a crucial point: while Generative AI technology isn’t novel, our recent capability to run Large Language Models (LLMs) efficiently owing to enhanced computational power is. Simply put, the monumental progression in AI development isn't primarily because of new mathematical models or scientific principles, but rather our refined capability to implement these models effectively. OpenAI's introduction of GPT APIs has played a pivotal role in this acceleration. As Professor Marwala aptly put it, even with the technology now freely available to run LLMs, not everyone has the infrastructure to harness LLM technology that require multiple large GPU arrangements. Therefore, the significance of accessing cutting-edge LLMs through an API is immense. This access has paved the way for an explosion of AI-centric products, including the impressive . MobileGPT Revolutionizing Through Refinement One might wonder why I’ve delved deep into this background information. It's crucial to understand that groundbreaking changes don’t always arise from entirely new inventions or groundbreaking innovations. More often, they stem from refining existing systems, enhancing processes, and expanding accessibility. It’s an eye-opening realization: the current AI surge, or as some critics label it, the "AI bubble," is powered more by accessibility and inclusivity rather than pure innovation. MobileGPT: Bridging the Gap Enter MobileGPT, a beacon in this era of AI accessibility. By integrating Generative AI technology with WhatsApp, it doesn't claim to reinvent the wheel. Instead, it focuses on magnifying accessibility, simplifying user experience, and fostering inclusivity for a vast audience. Considering that WhatsApp boasts over 2 billion users worldwide, the impact of cannot be overlooked. MobileGPT But what does this mean for an average user? For an average user, the introduction of MobileGPT is asking to unlocking a treasure trove of possibilities right within their most familiar communication tool. No longer is AI confined to the domains of tech-savvy enthusiasts or professionals with high-end computational resources. With MobileGPT integrated into WhatsApp, AI becomes as accessible as sending a text message. Imagine looking for specific information from a Wikipedia page or a blog. You can use our “chat to website” feature and enter the website link, let MobileGPT grab all the data from that website, add it to an index that you can now query. You can ask for specific insights, summaries or even extract points and contact information from websites using this feature. MobileGPT: Beyond the Ordinary Chatbot In the realm of chatbots, MobileGPT stands tall, not just as another conversational AI, but as a multifaceted tool that redefines user interaction. This powerful integration does more than merely employ Generative AI's capabilities such as instant chat, language translation, image creation, document generation, or swift online research. It serves as a personal digital companion, ready to assist in varied tasks, revolutionizing the way we interact with technology. The concept of jotting down notes might seem mundane, but its significance is undoubted. Most of us are swamped with myriad note-taking tools – from sophisticated apps on our smartphones to dedicated programs on our laptops or integrated systems in CRMs. Yet, faced with the simple task of noting down grocery items, many revert to the age-old trick of sending a WhatsApp message to oneself. The reason? Convenience. WhatsApp sits comfortably in our daily routines, making it the immediate go-to platform when we wish to jot something down quickly without the hassle of toggling through multiple apps. Understanding this innate user behavior, MobileGPT doesn’t seek to reinvent the wheel. Instead, it capitalizes on what users already love and trust, enhancing the experience. Why bother switching between multiple platforms when you can seamlessly integrate note-taking within your most-used messaging app? With MobileGPT, you're not limited to just one-off notes; you can store and manage up to 20 notes, ensuring they're always at your fingertips whenever you need them. Engaging with MobileGPT's note-taking functionality is as intuitive as conversing with a friend. Simple prompts like “save this note” or “show me my notes” suffice, making the entire experience fluid and hassle-free. Technology Inclusion Beyond the utility aspect, there's an even deeper significance. For many, technology can often feel overwhelming or alienating. By nestling AI within the confines of WhatsApp, MobileGPT demystifies the world of artificial intelligence. It offers an avenue where users, regardless of their tech proficiency, can engage with AI in a manner that feels intuitive, friendly, and uncomplicated. In essence, MobileGPT is not just about AI integration; it's about changing perceptions, breaking barriers, and ensuring that the wonders of AI are within reach of everyone. In a digital age that thrives on connection, this bridging of the AI accessibility gap is both timely and transformative. Wait, there is more MobileGPT recently introduced ‘magic commands’ - short cuts to simple tasks that one might need to undertake on a daily basis. One can now do the following with magic commands: 👉🏽 ekse google ___ (To do a quick google search. eg. ) ekse google womens world cup 2023 👉🏽 ekse jobs ___ (To do a job search. eg. ) ekse jobs remote python developer 👉🏽 ekse shopping ___ (For shopping searches. eg. ) ekse shopping New Samsung Flip Phone 👉🏽 ekse books ___ (To do a book search. eg. ) ekse books art of war 👉🏽 ekse link __ (Analyse a website. eg. ) ekse link https://mobile-gpt.io 👉🏽 ekse news __ (Get latest news. eg. ) ekse news business* (*choose 'business' or 'science' or 'sports' or 'entertainment' or 'health' or 'technology' Get Started Get started by visiting the website: https://mobile-gpt.io

Take Generative AI With You on the Go With MobileGPT WhatsApp Version

Lleve la IA generativa consigo mientras viaja con la versión MobileGPT de WhatsApp

Emportez l'IA générative avec vous lors de vos déplacements avec la version MobileGPT WhatsApp

MobileGPT व्हाट्सएप संस्करण के साथ चलते-फिरते जेनरेटिव AI अपने साथ रखें

MobileGPT WhatsApp バージョンを使用して、外出先でも生成 AI を持ち運べます

Leve IA generativa com você em qualquer lugar com a versão MobileGPT WhatsApp

Mang theo AI sáng tạo khi di chuyển với phiên bản WhatsApp MobileGPT

使用 MobileGPT WhatsApp 版本随身携带生成式 AI

"How hard can it be?" This is what I told myself the moment I decided to start learning how to write code from scratch. I went from not being able to write "hello world" in any programming language to building that was one of the first to bring ChatGPT to WhatsApp. MobileGPT - the WhatsApp application When I first decided to learn how to code, I was an engineer who had just left my job and was trying to start up a bitcoin exchange in 2017. I weighed all my options and considered: Hiring a developer to help me build the exchange (Way too expensive) Hiring a dev company (Even more expensive) Purchasing a white label exchange (I tried that, but the code was unreliable and inflexible) Learn how to code (Now that was an idea) In the end, I went with option 4. In this article, I will share my story and journey, what I learned, and how I learned it. Hopefully, it will help a few people considering coding as a new career path and also show others that it is possible to learn coding and go on to building amazing products. Setting Yourself Up for Success: Planning, Budgeting, and Clearing the Path for Learning How to Code Like any journey, you need to plan ahead. Most people who studied programming did so at school as part of their studies. There was a schedule, lecturers, assignments, and a routine. If you are switching careers, you might have to learn coding part-time while working, watching videos online when you get a chance. This is not going to be easy, but do not despair - it is completely possible and has been done before. I started by giving myself time. I created a schedule and a routine, assigning time in my day to do certain things consistently. I worked through the course content and built projects from the courses. The key here is consistency. When you decide to learn how to code, consistency will be important. Online courses will never be enough Sorry to burst that bubble, but online learning courses are just the foundation or starting point. They are never going to be enough if you want to become a real programmer. These are some of the places where I started learning: Udemy - https://www.udemy.com/ Udacity - https://www.udacity.com/ These two platforms have everything you need to get started, but ultimately, it was reverse engineering projects on GitHub that gave me the edge and knowledge to start building real-life projects. Reverse Engineering Projects This means finding code online for software built by professional developers, studying the code, and figuring out how they did it. At the time, I was learning blockchain development, so I was studying code created for . Yes, I was new to coding and thought blockchain development was the best way to get started. Hyperledger Sawtooth Today, I would not start with blockchain code. In fact, try working with code within your area of interest. I am into OpenAI APIs now, so a good place to start reverse engineering code would be OpenAI Cookbook - https://github.com/openai/openai-cookbook In my learning journey, watching videos and courses from Udemy and Udacity might have been 30% of my learning, 40% was reverse engineering code, and the remaining 30% was experience. You gain experience from building your own stuff, your own ideas. You take what you have learned from different places and stitch it together into your own unique project. My Programming Journey: Gaining Experience After I was convinced that I had learned enough from online courses and reverse engineering projects, I started building my own. I started with a mobile application on Android that would call the CoinMarketCap API for the latest crypto prices. I could input my portfolio of coins, and it could calculate the full value of my portfolio in real-time. It was a useful app that helped me realize one important thing: I never wanted to build another native mobile application again. It was just too tedious. Writing the same code for different operating systems and screen sizes was overwhelming. Even creating app icons required multiple files for different resolutions. As an engineer at heart, I did not have time for such tediousness. I knew I had to be a creator of ideas. Back-end development was an obvious choice. I would build back-end code using Python and work with systems that made sense to me. I started building web applications in Python with frameworks like Flask and Django. I liked the flexibility of Python, and very quickly, I was turning around projects faster than the industry norm. I used my engineering mind to find the path of least resistance and apply the theory of constraints to solve problems through code. I'm glad I didn't learn the "dev-house" way of building things. I simply imagined solutions and used code as a tool. So, of course, when ChatGPT came out, it was pretty obvious to me. I needed this on my phone, but not as another mobile app. I needed a better solution. I decided to integrate it with the WhatsApp Meta API and create an interface on WhatsApp. But how could I create enough complexity in WhatsApp to handle user profiles, payment processing, rate limiting, and more? MobileGPT: Implementing the OpenAI API on WhatsApp: When ChatGPT was introduced, I knew I needed it on my phone. However, I sought a more efficient solution than creating yet another mobile app. That's when the idea struck: integrating ChatGPT with the WhatsApp Meta API and building an interface within WhatsApp. I was on my way to building MobileGPT - the world's first ChatGPT implementation on WhatsApp, which was launched on Product Hunt on March 13th, 2023. But MobileGPT was not just another wrapper around ChatGPT. I added more features to make the product valuable to anyone seeking Generative AI capabilities on the go. These included: Creating AI-Powered Documents: With MobileGPT, you can easily create professional documents by providing a simple prompt and description. For example, if you need to develop a business plan, instead of spending hours on it, you can simply provide MobileGPT with basic information about your business idea, and it will generate a full business plan for you in less than a minute. Online Research: MobileGPT allows you to conduct real-time research on the web using AI. It can collect all the information from the first page of Google search results and prepare a comprehensive research report for any query. Just provide the query, and MobileGPT will do the rest. Creating AI Images: Generative AI and image generation work hand in hand, and now you can generate images directly within MobileGPT while having normal chat conversations with AI. MobileGPT has integrated Stable Diffusion XL for better image quality and output. Talk2PDF - Read a PDF Document: Upload a PDF document to MobileGPT and chat with it. Whether you need a summary, want to extract important insights, or simply understand the document without reading all 100 pages, MobileGPT has you covered. It can provide summaries and extract relevant information from the document, saving you time and effort. Reminders and Notes: MobileGPT allows you to easily create reminders and notes. Simply tell MobileGPT to save a specific message as a note or set a reminder for a particular task, and it will create and store the note or send you a WhatsApp reminder at the specified time. Notes are useful for saving important responses or information for future reference. Website Summary: With MobileGPT, understanding a website becomes effortless. Just provide MobileGPT with the website link, and it will summarize the content for you in a Word document. You can also chat with a website by using the "chat to website" feature, similar to how you would chat with a PDF. MobileGPT on WhatsApp offers an array of powerful features that extend beyond traditional ChatGPT capabilities, making it a versatile tool for various tasks. Check out MobileGPT here: https://mobile-gpt.io/

From Zero to Hero: Learning How to Code and Building MobileGPT

De cero a héroe: aprender a programar y construir MobileGPT

De zéro à héros : apprendre à coder et créer MobileGPT

ज़ीरो से हीरो तक: मोबाइलजीपीटी को कोड करना और बनाना सीखना

ゼロからヒーローへ: コードの学習と MobileGPT の構築方法

Do Zero ao Herói: Aprendendo a Programar e Construindo o MobileGPT

From Zero to Hero: Học cách viết mã và xây dựng MobileGPT

从零到英雄：学习如何编码和构建 MobileGPT

Read this article if you are excited about Generative Artificial Intelligence (Generative AI), ChatGPT, Stable Diffusion, and MobileGPT. In this article, we are going to explore the art of prompting, or prompt engineering. By the end of this story, you will be a master in generating prompts for Generative AI - we will cover both text and image-generating prompts. This article is for you if you want to maximize the output you get from the GenerativeAI tools you are using. What Is Prompt Engineering? Prompt engineering, at its most fundamental level, is the process of designing, crafting, and refining inputs or "prompts" that are used to guide an artificial intelligence (AI) model's outputs. The term is especially relevant when we're dealing with generative AI models, such as the ones developed by OpenAI (GPT-3 and GPT-4) or other similar technologies. Imagine you're giving instructions to someone to perform a task they have never done before. The clarity and specificity of your instructions can greatly influence the end result. Similarly, with AI, the way you frame your prompts can significantly affect the AI's responses. Step-by-Step Guide to Building Effective GenerativeAI Prompts for Text Generation There is no better way to learn than with examples; we will use - GenerativeAI on WhatsApp to demonstrate these concepts. MobileGPT Example Prompt: Write a Speech This is the response we get back from MobileGPT, the instruction is too vague for the AI to understand, and it will even respond back with more clarification questions. At a minimum, we need to provide a topic or subject. This brings us to the basics of prompting - you need to provide both the format and topic. : This is what the AI must return or submit back to you, it is the basis of your prompt. For instance, do you ask the AI to generate text in the form of a bullet-point list, a formal letter, a speech, or a conversation? Format : What is your format about? The subject, topic, or title of the content that forms the basis of the content that the AI will return. Subject In our example, we could say: Write a speech about ‘The dangers of Plagiarism’ The Title or Subject is: Dangers of Plagiarism Generative AI is so intelligent and well-trained that - Format and Subject alone can be enough to generate real great content. However, we want to refine this a little more, so I am going to introduce a few more variables that will customize this speech, exactly the way we want it. Consider: Tone Audience Objective Act As The Tone The refers to the general character, attitude, or sentiment conveyed by the content that the AI generates. It is the overall feeling or mood that is portrayed through the language used in the output. tone You can think of tone as a directive on how the AI should "sound" in its response. The Audience The refers to the intended recipients or consumers of the AI-generated content. Essentially, it is the group of people that the content is designed to reach or influence. audience Clarifying the audience is very important in GenerativeAI - delivering our speech to kindergarten class would not be the same as graduating class at Harvard for example - even if the subject was the same. Objective The objective refers to the intended purpose or goal of the AI-generated content. It's essentially what you aim to achieve with the content that the AI produces. Defining the objective can help AI pick relevant talking points for your content. Act As Defining "Act As" in a prompt is a way of instructing the AI to assume a specific role, perspective, or persona while generating the content. This can significantly influence the tone, style, language, and overall nature of the AI's output. In most cases, you will not need to define all these variables in one prompt. For example, if you choose to say: Act as a lecturer, you might not need to define a tone because the persona comes with a context and tone. Let us put these to the test and refine our prompt. Variation 1: Write a speech about ‘The dangers of Plagiarism’ to be delivered to high school graduates by Oprah Winfrey Variation 2: Write a speech about ‘The dangers of Plagiarism’ in a humorous tone to convince pre-school children about the dangers of plagiarism Changing the audience results in a completely different speech, specifying the tone helps the AI understand how to structure the content, and most importantly - defining the objective gives the AI specifications of what the content should achieve. MobileGPT Cheat Sheet I developed a summary reference guide, download it, share it, and when in doubt, refer back to the cheat sheet: But wait - there is more. You can customize the output you get from Generative AI by specifying more variables such as (1) background information, (2) keywords, (3) word count, (4) guidelines, and (5) outline. Background Information Say you are giving a speech, but you want the AI to know more about the context. You could give it some background information. For example, you could add: “I am new to this school, this is my first year, and I graduated top of my class; I would like to sound humble in this speech.” Keywords I must say GPT3 and GPT4 engines, which we call for MobileGPT responses generally love to ignore instructions about keywords, but I love to include them, especially when writing content for blogs or social media. It’s usually important to specify the important terms and words that must come through in the content generated. Word Count This is another variable GPT3 and GPT4 engines love to ignore, but not always, depending on how you phrase it. It might be better to say: write me 5 paragraphs with 2 sentences each instead of saying: write me 500 words. Guidelines Honestly, “guidelines” alone, as an added variable can mean the difference between a generic response and a well-engineered AI output. Guidelines are like sets of instructions on “how” the AI should go about structuring your content. Imagine you want to write a research report and you tell AI: Write a research report on the impact of social media on mental health. You could add audience, tone, and objective, but you will probably need to give the AI more structure. This is where guidelines come in; I normally say: Write a research report on the impact of social media on mental health. --Start Guidelines-- Start by providing an executive summary highlighting all the important points and takeaways. Discuss the benefits and challenges in a critical manner, producing evidence for and against the topic. Give a brief background on the topic, discuss the literature around it, and present the different schools of thought. Discuss how the topic has evolved over the years and how the literature and schools of thought have shifted. Add a SWOT analysis towards the end and discuss the outcomes. Finally, summarize the key takeaways and present the findings in bullet points. --End Guidelines-- Let us test this out on MobileGPT, this time we will use the “Long Research Report” functionality. Just click on the menu, from any message and choose - Long Research Report. Then enter the title of your report: The Impact of Social Media on Mental Health That is it: MobileGPT handles the prompt engineering in the background, collects real data from the internet, and returns a report in Word format. You will also get links used to compile the data - so you can go expand on the report in your own time. Outline It sometimes helps to give Generative AI content outline, especially if you are generating content for reports, blogs, or content that needs to follow a prescribed outline. I will use blog writing as an example, you could provide an outline like this: Generate a blog about the dangers of plagiarism, and include the following subheadings in the blog. (1) Introduction - discuss the purpose of the blog, (2) Benefits - Talks about the benefits of plagiarism, (3) Discuss the challenges with plagiarism, (4) Conclude - write a compelling conclusion. Building Effective Prompts for Image Generation With AI AI image generation might be easier as there are fewer things and variables you have to remember. As we did with text generation, we will start with simple prompts, and build up the complexity with more detail to refine the output and customize what we will get from the AI. We will work with the prompt: Generate an image of a dog Lucky for us, we can also do this on MobileGPT - select Image Generation from the menu, then enter your prompt. This is not the dog I wanted. When it comes to image generation, you really have to think like an artist. Start visualizing this dog, what type of dog is it, what is it doing, and where is it. Define the subject in detail. Define what the subject is doing Define the background image in detail. Let us try this again: Generate an image of a white chihuahua, running on the pavement, in front of a house, surrounded by blue skies and tall trees. The image looks much better, but there is more. Consider adding the image type, you can choose from: Image Type Photorealistic: Images that look like photographs Sketch: Pencil or charcoal sketches Watercolor images Oil Painting images Pointillism: The image is made up of small dots of color. They have a unique look and feel, with a lot of detail and texture. Cartoon: These are images that are generated to look like cartoons. Anime/Manga: Images that look like Japanese anime or manga. Pop Art: Images that look like pop art, with bold colors and graphic shapes. Background Settings An often overlooked variables when doing image generation is the setting. This usually goes hand in hand with the image type and depends on the subject of the image. For example, it makes sense for a model to be in a studio setting, perhaps it’s a photo shoot, and for a bird to be in a natural setting, like a forest. Here are some examples: AI image generation technologies have advanced to a point where they can create a wide range of backgrounds and settings, complete with complex lighting and textural details. The specific capabilities may depend on the particular AI model and its training data, but here are a few examples of the types of backgrounds and settings that could be created: : This could include a basic backdrop (like white, black, or a specific color) with controlled, professional lighting. Props and models can be included as well, with the AI adjusting their appearances to match the studio lighting conditions. Studio Setting : These could include forests, mountains, beaches, deserts, oceans, and other landscapes. Lighting conditions could vary from bright sunlight to the soft hues of a sunset or the cool tones of a moonlit night. Natural Settings : Cityscapes with skyscrapers, busy streets, parks, or suburban neighborhoods are all possible. Again, lighting conditions can be varied to simulate different times of day or weather conditions. Urban Environments : These could be home interiors, offices, restaurants, schools, museums, or any other type of indoor setting. Lighting can be adjusted to match typical indoor conditions like the warm glow of a table lamp or the bright fluorescent lights of an office. Indoor Scenes : Depending on the AI's training data, it might be possible to generate images of historical settings (like a medieval village or a 1920s city street) or fictional environments from books, movies, or games. Historical or Fictional Settings : The AI can generate purely abstract or artistic backgrounds with a variety of colors, shapes, and patterns. Abstract Backgrounds Let us apply what we have learned now and create a futuristic car fit for a magazine cover: Generate a photo-realistic image of a futuristic electric BMW car, in a studio setting with perfect colour and light shinning on the metallic surface of the car. And this was just my first attempt; you can refine the prompt by adding how the car is positioned, different camera angles, etc. Hope you learned something; feel free to check out MobileGPT here: Add it to your WhatsApp experience, and become a prompt master. https://mobile-gpt.io/

Your Ultimate Guide to Mastering AI Prompt Engineering With MobileGPT

Su guía definitiva para dominar la ingeniería rápida de IA con MobileGPT

Votre guide ultime pour maîtriser l'ingénierie rapide de l'IA avec MobileGPT

MobileGPT के साथ AI प्रॉम्प्ट इंजीनियरिंग में महारत हासिल करने के लिए आपकी अंतिम मार्गदर्शिका

MobileGPT を使用して AI プロンプト エンジニアリングをマスターするための究極のガイド

Seu guia definitivo para dominar a engenharia de prompt de IA com o MobileGPT

Hướng dẫn cơ bản của bạn để làm chủ kỹ thuật AI Prompt với MobileGPT

使用 MobileGPT 掌握 AI 快速工程的终极指南

Say hello to MobileGPT, a game-changing startup that is revolutionizing your WhatsApp experience by integrating the powerful OpenAI models, GPT3.5-turbo and GPT-4, directly into your chats. MobileGPT is AI at your fingertips, providing AI-enhanced document creation, image generation, conversation, and even deep dives into PDFs and websites, all within WhatsApp. We’ve been nominated for HackerNoon’s Startups of the Year and this is our interview: Vote for us today! MobileGPT Voting Page My Role Hello, I am Bertha Kgokong, a full-time entrepreneur and software developer with a passion for designing impactful software. One such innovation is MobileGPT, where I have harnessed my expertise to create a tool that brings AI to your mobile device in a practical, accessible, and user-friendly manner. Besides MobileGPT, I'm also behind the creation of , a versatile business management software. Skhokho.io How We're Disrupting the Tech Industry MobileGPT is on the path to radically disrupt the tech industry by embedding the powerful capabilities of GPT AI models into one of the most popular messaging apps worldwide - WhatsApp. We're transforming a simple chat platform into a comprehensive productivity tool, equipped with functionalities like AI document generation, talk to PDF features, creating reminders and notes, and even website summarization. By making these cutting-edge AI capabilities accessible to millions of WhatsApp users, we're effectively democratizing the power of AI. Standing Out from The Crowd While there are a myriad of AI tools out there, what sets MobileGPT apart is its integration with a platform that is widely used and familiar to many - WhatsApp. We're taking the complexity out of AI and embedding it in your daily routines. Need a summary of a long research report? Just use MobileGPT's LiveData function. Want a visually stunning image based on a text description? MobileGPT can handle that. All of these features are provided in an interface you already know and love, removing the learning curve that often comes with new tech tools. Our Predictions on the AI Tech Industry in 2023 As we move further into 2023, we see AI continuing to break barriers and becoming even more integral in our everyday lives. Its accessibility and usability will continue to improve, becoming more embedded in common applications and software we use daily. We also anticipate an increased focus on ethical AI and responsible AI use. AI's potential in education is particularly exciting. With tools like MobileGPT, we foresee a shift towards more personalized and adaptive learning experiences, powered by AI. What word defines the state of Tech in 2023? Integration. The tech industry in 2023 is about integrating sophisticated technologies such as AI into everyday applications and platforms to enhance usability, productivity, and overall user experience. It's about making AI ubiquitous and seamlessly woven into our daily digital interactions. Why we decided to participate in HackerNoon's Startup of the Year awards We chose to participate in HackerNoon's Startup of the Year awards because we believe in the power of our product and the positive impact it can have on people's lives. We want to showcase how we're using AI to create user-friendly, practical tools that boost productivity and transform everyday interactions. Final Thoughts At MobileGPT, our mission is to empower users by bringing the power of AI directly to their fingertips. By integrating these capabilities into familiar platforms like WhatsApp, we're breaking down barriers to AI adoption and making this powerful technology more accessible than ever. As we continue to innovate and grow, we're excited to see how AI continues to evolve and reshape our digital landscape. Vote for us today! MobileGPT Voting Page

Meet MobileGPT: Nominated for HN's Startups of the Year

Conozca MobileGPT: Nominado a Startups del año de HN

Rencontrez MobileGPT : nominé pour les startups de l'année de HN

मोबाइलजीपीटी से मिलें: एचएन के वर्ष के स्टार्टअप के लिए नामांकित

MobileGPT のご紹介: HN の Startups of the Year にノミネート

Conheça o MobileGPT: indicado para Startups do ano da HN

Gặp gỡ MobileGPT: Được đề cử cho Startups of the HN

认识 MobileGPT：HN 年度初创公司提名

After processing around 30,000 startups and more than 600,000 votes, the curtain has closed on . Thank you all for changing the world one product/service at a time. Startups Of The Year 2023 We’re excited to announce the Startups of the Year across Africa! Africa is a continent of vast diversity, home to 54 countries with rich cultures, landscapes, and a rapidly evolving economy. In recent years, Africa has emerged as a dynamic player in the global startup ecosystem, showcasing a vibrant and entrepreneurial spirit. The continent is experiencing a surge in tech-driven initiatives, fueled by a young and growing population, increased access to technology, and a rising middle class. Key Stats Total Number of Participating Startups: 831 Total Number of Cities Represented: 129 Total Number of Votes cast: 13696 Winning Startups: 71 To see who is a winner in each respective city, visit and pick a city. The name at the top of the list is the winner. Note: startups.hackernoon.com Now, drumroll, please!!! Here are the Startups of the Year in Africa Enugu, Nigeria The Startup of the Year in Enugu is Superday with 57 votes! 🎉 Superday provides accountability tools for online learning. Addis Ababa, Ethiopia This year’s top Startup in Addis Ababa is Sabi with 28 votes! 🎉 Sabi provides affordable and vetted tech talent sourced from Africa to startups and businesses. Algiers, Algeria Go Platform emerged as Startup of the Year in Algiers with 8179 votes!🎉 GO PLATFORM is a Recruitment Process Outsourcing SaaS solution, it’s a two-sided marketplace for digital talent. Ondo, Nigeria Beepo is Startup of the Year in Ondo with 76 votes!🎉 Beepo offers intuitive solutions to Africa’s growing blockchain industry. Ariana, Ariana Governorate, Tunisia The Startup of the Year in Ariana is Work Point with 4 votes!🎉 Workpoint offers solutions in the HRTech Industry Cairo, Egypt AbuEran is the Startup of the Year in Cairo after getting 34 votes!🎉 AbuEran is a poultry management solution that empowers poultry companies with AI to maximize production and reduce waste. Uyo, Akwa Ibom, Nigeria LINK IO is the top Startup in Uyo with 4 votes!🎉 LINK allows users around the world to receive cash fast and cheaply from Africa. Accra, Ghana Tudu claims the Startup crown in Accra with 767 votes!🎉 Tudu helps online sellers and buyers transact with confidence. Lagos, Nigeria Treford is the king of the hill in Lagos with 415 votes!🎉 Treford is an EdTech startup focused on helping professionals and businesses in Africa gain no-code tech skills. Buea, Southwest, Cameroon The Startup of the Year badge for Buea, Southwest, Cameroon goes to Klarah with 4 votes!🎉 Klarah is in the Healthcare IT Industry. Kingston, St Andrew Parish, Jamaica HummingBird is the top Startup in Kingston, St Andrew Parish, Jamaica after garnering 4 votes!🎉 HummingBird is in the HRTech Industry Dar es Salaam, Tanzania KaziConnect is the top Startup in Dar es Salaam after garnering 5 votes!🎉 KaziConnect is in the HRTech Industry. Cape Town, South Africa Dimagi is Cape Town’s Startup of the Year with 31 votes!🎉 Dimagi is a social enterprise that powers impactful frontline work through scalable digital solutions and services. Lekki, Nigeria Lekki’s Startup of the Year is Eazipay with 4 votes!🎉 Eazipay provides solutions in the HRTech Industry. Kano, Nigeria In Kano, Insigning is the Startup of the Year with 4 votes!🎉 Insigning offers services in the MarketingTech Industry. Giza, Egypt The Startup of the Year in Giza is CirclePay with 13 votes!🎉 CirclePay provides services in the Payments Industry Minna, Niger, Nigeria Timart Business App is the Startup of the Year in Minna with 2 votes!🎉 Timart Business App serves the Enterprise Resource Planning Industry. Victoria, Seychelles The Startup of the Year in Victoria is StorX with 4 votes!🎉 StorX does exciting work in the Enterprise Storage Industry. Agadir, Souss Massa, Morocco Atela is the top Startup in Agadir, Souss Massa with 4 votes!🎉 Atela services the Investment Tech Industry. Casablanca, Morocco Invoiless is the top Startup in Casablanca, Morroco with 15 votes!🎉 Invoiless is in the Finance-Accounting Tech Industry Free State, South Africa Africarare ended the year as the top Startup in Free State with 4 votes!🎉 Africarare is in the Virtual Reality Industry. Dakar, Senegal The Startup of the Year in Dakar is Paps with 4 votes!🎉 Paps provides services in the Logistics Tech Industry. Ogbomosho, Oyo, Nigeria Phalconwise is the Startup of the Year in Ogbomosho with 265 votes!🎉 Phalconwise offers Banking Tech services. Johannesburg, South Africa The Startup of the Year in Johannesburg is MobileGPT with 127 votes! 🎉 MobileGPT is an innovative AI-driven service designed to enhance your WhatsApp experience by seamlessly integrating the cutting-edge GPT technology from OpenAI. Abuja, Nigeria Our Haven emerged as Startup of the Year in Abuja with 119 votes! 🎉 Our Haven is a tech-enabled platform that allows customers to pass on their tedious everyday chores and household needs to us while they focus on more important things. Nairobi, Kenya The top Startup in Nairobi is ProStudy, amassing 90 votes! 🎉 Prostudy provides services in the K-12 EdTech Industry. Ikeja, Nigeria Legalforms.ng is the Startup of the Year in Ikeja with 17 votes! 🎉 Legalforms.ng is in the Legal Tech Industry. Harare, Zimbabwe YoApp ended the year as the top Startup in Harare with 13 votes! 🎉 YoApp is in the Customer Success Management Industry. Durban, KwaZulu Natal, South Africa The Startup of the Year in Durban is Nova DT with 11 votes! 🎉 Nova DT shines in the Logistics Tech Industry. Ikorodu, Lagos, Nigeria Ikorodu’s Startup of the Year is Infonaira with 11 votes! 🎉 Infonaira provides services in the HRTech Industry. Oran, Oran Province, Algeria In Oran, Scan Docs is the Startup of the Year with 11 votes! 🎉 Scan Docs is in the Digital Transaction Management Industry. Asaba, Delta, Nigeria The Startup of the Year in Asaba is Contactless Biz Card with 4 votes!🎉 Contactless Biz Card is in the Sales Force Automation Industry. Federal Capital Territory, Nigeria Envoicify emerged as the Startup of the Year in the FCT with 4 votes! 🎉 Envoicify is in the SCM Software Industry. Marrakesh, Marrakesh-Safi, Morocco The Startup of the Year in Marrakesh is Tidar with 4 votes! 🎉 Tidar is in the Events Tech Industry. Mogadishu, Banaadir, Somalia Mogadishu’s Startup of the Year is Hassildesk with 4 votes!🎉 Hassildesk is in the Online Travel Industry. Stellenbosch, Western Cape, South Africa Scout is Startup of the Year in Stellenbosch with 4 votes! 🎉 Scout is in the MarketingTech Industry. Tema, Greater Accra Region, Ghana Tema crowns Nulave Startup of the Year with 4 votes! 🎉 Nulave is in the Local Services Industry. Western Cape, South Africa The Startup of the Year in Western Cape is Mobitrax with 4 votes!🎉 Mobitrax is in the IoT Applications Industry. Port Harcourt, Nigeria Secentry is the Startup of the Year in Port Harcourt with 3 votes! 🎉 Secentry is in the Facility Management Tech Industry. Tangier, Tanger Tetouan Al Hoceima, Morocco ALLO GPS is the Startup of the Year in Tangier with 3 votes! 🎉 ALLO GPS offers services in the Logistics Tech Industry. Greater Accra Region, Ghana MakeWeHelp takes home the Gold in the Greater Accra Region with 2 votes!🎉 MakeWeHelp provides services in the Nonprofit Tech Industry. Ibadan, Oyo, Nigeria In Ibadan, HealthStack Solution is top of the pile with 2 votes!🎉 HealthStack Solution is in the Healthcare IT Industry Ilorin, Kwara, Nigeria EDX is the top Startup in Illorin with 2 votes! 🎉 EDX pioneers in the Healthcare IT Industry. Kaduna, Nigeria Global Care takes the cake in Kaduna with 2 votes!🎉 Global Care is in the Healthcare IT Industry Lome, Maritime Region, Togo Of all the startups in Lome, Rafset Funding came out top with 2 votes!🎉 Rafset Funding is in the Nonprofit Tech Industry. Nairobi County, Kenya Book Kinda beat other startups in Nairobi County for the crown with 2 votes!🎉 Book Kinda is in the SME Business Management Software Industry. Osogbo, Osun, Nigeria FinPadi scoops the victory with Osogbo with 2 votes!🎉 FinPadi services the Payments Industry. Tunis, Tunisia The Startup of the Year in Tunis is Bravvo with 2 votes!🎉 Bravvo is in the Enterprise Collaboration Industry. Freetown, Western Area, Sierra Leone Smart Systems stands at the top of the podium in Freetown with 6 votes!🎉 Smart Systems offers services in the Payments Industry. Gauteng, South Africa Gauteng dubs faceATT Startup of the Year with 39 votes!🎉 faceATT provides services in the Computer Vision Industry. Lusaka, Lusaka Province, Zambia Zipangile is the Startup of the Year in Lusaka with 24 votes!🎉 Zipangile works in the Application Development Tools Industry. Douala, Littoral, Cameroon Postpoint is the Startup of the Year in Douala with 12 votes!🎉 Postpoint helps you find, share, and accurately indicate a location with our innovative digital address solution. Congratulations y’all. We will contact you with details on your winnings! See the global and other regional announcements . here About HackerNoon’s Startups Of The Year Startups of The Year is HackerNoon’s Flagship community-driven event celebrating startups that survived and thrived in 2023. 30,000 startups across 4200+ cities and six continents participated in 2023 to be crowned the best startup in their city. Visit our page to learn more. FAQ Companies that blog more get 67% more leads - Jumpstart your company’s blogging journey by purchasing our . Business Blogging Package The 2023 Startups Of The Year is sponsored by: . . The winners will get a free .Tech Domain, a HackerNoon NFT, and a Tech Company News Page. TECH Domains

Startups of the Year 2023: Africa Winners

As someone always on the lookout for new and innovative ways to streamline my workflow, I’ve come across some incredible AI tools that have completely transformed the way I work. From chatbots that provide instant answers to tools that help me understand complex legal documents, these AI-powered solutions have made my life so much easier. In this article, I’ll be sharing some of the latest trending AI tools that you need to know about. Merlin Merlin is a tool that gives users the power of OpenAI ChatGPT Plus on all their favorite sites for free. With Merlin, users can supercharge their productivity by leveraging the power of AI to automate tasks and get more done in less time. D-ID D-ID is a tool that allows users to create real-time face animations that can create a human-like experience. This tool is ideal for those looking to add a personal touch to their communications, making them more engaging and memorable. There’s an AI for that Find AI tools easily. This website features over 2000 tools for over 600 tasks. It is sponsored by ChatGPT and is updated daily. If you’re wondering if your task can be automated, there’s probably an AI for that. Berri Berri is a platform that allows users to build production-ready ChatGPT apps in minutes. With its powerful AI capabilities and user-friendly interface, Berri is perfect for anyone looking to create custom chatbots that can help them automate tasks and get more done in less time. Lindy Lindy is an AI assistant that can help users with all kinds of personal tasks. With its advanced AI capabilities, Lindy can help users stay organized, manage their schedules, and get more done in less time. 2short 2short is a tool that allows users to elevate their YouTube content with AI-generated shorts. With 2short, users can create engaging and attention-grabbing shorts that can help them grow their audience and increase engagement. Promptitude Promptitude is the easiest way to integrate GPT into your apps. With its powerful AI capabilities and user-friendly interface, Promptitude is perfect for anyone looking to add AI capabilities to their apps and supercharge their productivity. Adobe Firefly Adobe Firefly is a tool that allows users to make an infinite range of image creations. With its powerful AI capabilities and user-friendly interface, Adobe Firefly is perfect for anyone looking to create custom images that can help them tell their story and engage their audience. Google Bard Google Bard is Google’s new chatbot that takes on ChatGPT. With its advanced AI capabilities, Google Bard can help users automate tasks and get more done in less time. ChatBotKit ChatBotKit is the fastest way to make your own advanced chatbot. With its powerful AI capabilities and user-friendly interface, ChatBotKit is perfect for anyone looking to automate tasks and streamline their workflows. Superhuman Superhuman is a tool that can automatically sort important emails based on behavioral patterns. With its advanced AI capabilities, Superhuman can help users stay organized and get more done in less time. Venturus Venturus is a tool that allows users to get instant feedback and analysis of their business ideas. With its powerful AI capabilities and user-friendly interface, Venturus is perfect for anyone looking to make data-driven decisions and grow their business. Clip Fm Clip Fm is a tool that allows users to turn their videos into viral shorts with just one click. With its powerful AI capabilities and user-friendly interface, Clip Fm is perfect for anyone looking to grow their audience and increase engagement. ChatDOC This AI-powered tool allows users to chat with their documents and receive instant answers to their questions. This can save valuable time for those who frequently work with large amounts of documentation. Vizologi Business owners and entrepreneurs can benefit from Vizologi, which provides answers to all their business questions. From market analysis to competitor research, this AI tool can provide valuable insights to help businesses succeed. ContractReader Smart contracts can be difficult to understand, but ContractReader uses AI to make the process easier. By analyzing and interpreting contracts, ContractReader can provide users with clear explanations of their contents. Cody ChatGPT can be trained to learn about a business, team, or process using Cody. By inputting information, Cody can help create a personalized AI assistant that can improve productivity and streamline operations. WTF Does This Company Do? This tool helps users quickly understand the purpose of a company’s landing page. By simply entering the domain, users can receive an explanation of the company’s services and products. ProductBot This AI-powered tool acts as a product expert and recommender, helping users make informed purchasing decisions. By providing personalized recommendations, ProductBot can improve the shopping experience for users. Monica ChatGPT-4 powers Monica, an AI assistant that can help users with any website they visit. By providing quick answers to questions and helping with tasks, Monica can improve productivity and efficiency. MobileGPT ChatGPT can now be accessed on Whatsapp through MobileGPT. This can make it easier for users to access the AI assistant on the go. AI Course Creator This tool allows users to create complete courses using AI. By providing personalized content and assessments, users can create unique courses that meet their specific needs. AI is evolving quickly and offers powerful tools for developers and businesses. The examples discussed in this article can revolutionize problem-solving and decision-making. Machine learning algorithms, natural language processing, and computer vision can streamline operations, automate tasks, and provide insights from data. Businesses need to stay current with AI advancements to use technology effectively and remain competitive. I hope this list was helpful, feel free to comment if you have any more! Here are some creators that helped me put together this list: Barsee A lex Banks Zain Kahn Feature Photo by on Arseny Togulev Unsplash

22 AI Tools You Should Know About

The objective of the problem is to implement classification and localization algorithms to achieve high object classification and labelling accuracies, and train models readily with as least data and time as possible. The solution to the problem is considered in the following blog. The efficiency of a model is dependent on various parameters, including the architecture of the model, number of weight parameters in the model, number of images the net has been trained on, and the computational power available to test the models in real time. The third parameter can’t be controlled, thus leaving us dependent on the first two parameters. Thus transfer learning works the best in this scenario, for the pre-trained weights are adjusted according to our dataset, although minimal errors and reliable accuracies are obtained. Model For image classification, we use a keras model with the model summary obtained by running the code below The model’s parameters are tuned to suit the maximum change in information for as minimum data as possible. Thus, we have the batch normalization layers, that randomly shake up the weights to make the model generalized. . Experiment We use the MobileNet model for training on our dataset. The dataset has been taken from HackerEarth deep learning challenge to classify animals. If you need any other domain-specific dataset: . You can find thousands of such open datasets here We choose 10 random classes from the dataset and change the number of images per class and the size of the fully connected layers, and report the results. keras applications keras.preprocessing.image ImageDataGenerator keras optimizers keras.models Sequential, Model keras.layers Dropout, Flatten, Dense, GlobalAveragePooling2D keras backend k keras.callbacks ModelCheckpoint, LearningRateScheduler, TensorBoard, EarlyStopping keras.models load_model os pickle keras.models model_from_json matplotlib.pyplot plt

image_width, image_height= , nb_train_samples= nb_validation_sample= batch_size = model = applications.mobilenetv2(weights= , include_top=False, input_shape=(image_height, image_width, ))

x=model.layers[ ].output
#take the first layers the model
x=Flatten()(x)
x=Dense( , activation= )(x)
x=Dropout( )(x)
x=Dense( , activation= )(x)
x=Dropout( )(x)
x=Dense( , activation= )(x)
x=Dropout( )(x)
predictions = Dense( , activation= )(x)


model_final =Model(input=model.input, output=predictions)

model_final = load_model( )
model_final.compile(loss= , optimizer=optimizers.nadam(lr= ), metrics=[ ])

train_datagen = ImageDataGenerator(rescale = / ,
                                   shear_range = ,
                                   zoom_range = ,
                                   horizontal_flip = True,
                                   fill_mode= ,
                                   width_shift_range= ,
                                   height_shift_range= ,
                                   rotation_range= )

test_datagen = ImageDataGenerator(rescale = / ,
horizontal_flip = True,
fill_mode = ,
zoom_range = ,
width_shift_range = ,
height_shift_range= ,
rotation_range= )

training_set = train_datagen.flow_from_directory( , target_size = ( , ), batch_size = ,class_mode = )
test_set = test_datagen.flow_from_directory( , target_size = ( , ), batch_size = , class_mode = ) 
model_final.fit_generator(training_set, steps_per_epoch = ,epochs = , validation_data = test_set,validation_steps= )
print(model.summary())
#uncomment the follwoing to save your weights and model. model.json r categorical_crossentropy nadam accuracy from import from import from import from import from import from import as from import from import import import from import import as 256 256 11000 2000 8 "imagenet" 3 7 5 of 1024 "relu" 0.5 384 "relu" 0.5 96 "relu" 0.5 30 "softmax" "weights_Mobile_Net.h5" "categorical_crossentropy" 0.00001 "accuracy" 1. 255 0.2 0.2 "nearest" 0.3 0.3 30 1. 255 "nearest" 0.3 0.3 0.3 30 './HE_Chal' 256 256 8 'categorical' './Validation' 256 256 8 'categorical' 1000 80 1000 '' 'model_json=model_final.to_json()

with open("model.json", "w") as json_file:
    json_file.write(model_json)
model_final.save_weights("weights_VGG.h5")
model_final.save("model_27.h5")
#model_final.predict(test_set, batch_size=batch_size)
' '' '' '
json_file = open(' ', ' ')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("weights_VGG.h5",by_name=True)
print("Loaded model from disk")
 
# evaluate loaded model on test data
loaded_model.compile(loss=' ', optimizer=' ', metrics=[' '])

#print(loaded_model.summary())
loaded_model.fit_generator(training_set,                         steps_per_epoch = 1000,epochs = 100,                         validation_data = test_set,validation_steps=1000)
#score = loaded_model.evaluate(training_set,test_set , verbose=0)
' '' Results The models were run for 15 epochs on an Intel i7 Processor. Model 1: Mobilenet : 1000 steps/epoch: Larger FC Layers: Training Time: 18 mins/epoch : dataset: 50 images/average: accuracy : 82.2% Model 2: Mobilenet: 500 steps/epoch: Smaller FC layers: Training time : 12 mins/epoch: dataset: 50 images/average : 82.47% Model 3: Mobilnet 500 steps/epoch: Smaller FC layers: Training time: 11 mins/epoch: dataset: 30 images per class: accuracy: 76% Image Detection There are a few methods that pose detection as a regression problem. Two of the most popular ones are YOLO and SSD. These detectors are also called single shot detectors. Let’s have a look at them: You Only Look Once. YOLO divides each image into a grid of S x S and each grid predicts N bounding boxes and confidence. The confidence reflects the accuracy of the bounding box and whether the bounding box actually contains an object(regardless of class). YOLO also predicts the classification score for each box for every class in training. You can combine both the classes to calculate the probability of each class being present in a predicted box. So, total SxSxN boxes are predicted. However, most of these boxes have low confidence scores and if we set a threshold say 30% confidence, we can remove most of them as shown in the example below. (YOLO predicts one type of class in one grid! Hence small objects are not identified…) Single Shot Detectors SSD runs a convolutional network on input image only once and calculates a feature map. Now, we run a small 3×3 sized convolutional kernel on this feature map to predict the bounding boxes and classification probability. SSD also uses anchor boxes at various aspect ratio similar to Faster-RCNN and learns the off-set rather than learning the box. In order to handle the scale, SSD predicts bounding boxes after multiple convolutional layers. Since each convolutional layer operates at a different scale, it is able to detect objects of various scales. We compared two models, initially YOLO(darknet) and later SSDs and compared their accuracies and speeds. Since our inputs are images, the FPS parameter is not used to differentiate the models. Moreover, the SSDs are a balance between the Faster — RCNN model and the YOLO model. Let’s see what the experiment tells us? The SSD model is implemented using the dnn module, with the help of Adrian Rosebrock, in openCV’s library. # the necessary packages numpy np argparse cv2

# construct the argument parse and parse the ap = argparse.ArgumentParser()
ap.add_argument( , , required=True,
	help= )
ap.add_argument( , , required=True,
	help= )
ap.add_argument( , , required=True,
	help= )
ap.add_argument( , , type=float, = ,
	help= )
args = vars(ap.parse_args())
# load our serialized model disk
print( )
net = cv2.dnn.readNetFromCaffe(args[ ], args[ ])

# load the input image and construct an input blob the image
# by resizing to a fixed x300 pixels and then normalizing it
# (note: normalization is done via the authors the MobileNet SSD
# implementation)
image = cv2.imread(args[ ])
(h, w) = image.shape[: ]
blob = cv2.dnn.blobFromImage(cv2.resize(image, ( , )), ,
	( , ), )
  
# pass the blob through the network and obtain the detections and
# predictions
print( )
net.setInput(blob)
detections = net.forward()

# loop over the detections i np.arange( , detections.shape[ ]):
	# extract the confidence (i.e., probability) associated the
	# prediction
	confidence = detections[ , , i, ]

	# filter out weak detections by ensuring the is
	# greater than the minimum confidence confidence > args[ ]:
		# extract the index the = int(detections[ , , i, ])
		box = detections[ , , i, : ] * np.array([w, h, w, h])
		(startX, startY, endX, endY) = box.astype( )

		# display the prediction
		label = .format(CLASSES[idx], confidence * )
		print( .format(label))
		cv2.rectangle(image, (startX, startY), (endX, endY),
			COLORS[idx], )
		y = startY - startY - > startY + cv2.putText(image, label, (startX, y),
			cv2.FONT_HERSHEY_SIMPLEX, , COLORS[idx], )
      
     # show the output image
cv2.imshow( , image)
cv2.waitKey( ) import import as import import arguments "-i" "--image" "path to input image" "-p" "--prototxt" "path to Caffe 'deploy' prototxt file" "-m" "--model" "path to Caffe pre-trained model" "-c" "--confidence" default 0.2 "minimum probability to filter weak detections" from "[INFO] loading model..." "prototxt" "model" for 300 of "image" 2 300 300 0.007843 300 300 127.5 "[INFO] computing object detections..." for in 0 2 with 0 0 2 `confidence` if "confidence" of ` `,
		# ( , )- # class label from the detections then compute the x y coordinates of the bounding box for the object idx 0 0 1 0 0 3 7 "int" "{}: {:.2f}%" 100 "[INFO] {}" 2 15 if 15 15 else 15 0.5 2 "Output" 0 The YOLO pre-trained weights were downloaded from the author’s website where we choose the YOLOv3 model. Since it is the darknet model, the anchor boxes are different from the one we have in our dataset. Hence we initially convert the bounding boxes from VOC form to the darknet form using code from . Then we train the network by changing the config file. here Results: SSDs: IOU= 0.74, mAP: 0.83 Time /epoch: 12 minutes YOLOs: IOU= 0.69, mAP: 0.85 Time/epoch: 11 minutes Output Images (SSDs used for Vehicle Detection) (Output with YOLOv3 Pretrained Weights) Conclusion The overall problem is stated as one where we need to trade off the speed and accuracy. The overall solution is proposed as two different models for various types of images. The trade-off between speed and accuracy is accompanied with computational power available. The YOLO model is suitable for high-speed outputs, where accuracy is not that high… whereas SSDs provide higher accuracies with high-speed outputs with a higher computation time. Hence choose SSDs on good microprocessors, else YOLO is the goto for microprocessor-based computations. Shameless plugin: We are an online tool to make it super easy for you to build ML datasets like doing image bounding boxes, NER tagging etc. Check us out: the best image labeling tool !

Efficient Implementation of MobileNet and YOLO Object Detection Algorithms for Image Annotation

In the first , we covered the two main aspects of deploying a deep learning model: part Preparing the dataset Training the model using the transfer learning technique. Part 2 will focus on preparing a trained model to be served by TensorFlow Serving and deploying the model to Heroku. TensorFlow Serving Is a flexible, high-performance serving system for machine learning models, designed for environments. I am not going to cover it in details because TensorFlow Serving has many ways to serve models and there are too many concepts to digest for this short tutorial. production I strongly encourage TensorFlow Serving beginners to first read, . However, this tutorial will cover how to serve Keras models and using the new RESTful API, which is not covered in that tutorial. TensorFlow Serving 101 pt. 1 Serve the model. First, let’s load the saved model from Part 1 from keras.models import load_model # model_name = tf_serving_keras_mobilenetv2model = load_model(f"models/{model_name}.h5") Save the Model We define the input and output tensors. The name and can be any string. Then we specify a path where the model will be saved, . is the version of the model; this is very useful when you are going to retrain the model and have a new version. _images_ _scores_ _image_ _score_ models/export/tf_serving_keras_mobilenetv2/1 _1_ Each version will be exported to a different sub-directory under the given path. Lastly, we build the model by passing the TensorFlow session, input, and output tensors. is the TensorFlow session that holds the trained model you are exporting [3]. sess is the set of tags with which to save the meta-graph. In this case, since we intend to use the graph in serving, we use the tag from predefined SavedModel tag constants [3]. tags serve specifies what type of model is being exported, and the input/output tensors to bind to when running inference. signature_def_map Containerize Your Model for Deployment — Docker What is Docker? Why Docker? is a tool designed to make it easier to create, deploy, and run applications by using containers. Containers allow a developer to package up an application with all of the parts it needs, such as libraries and other dependencies, and ship it all out as one package. By doing so, thanks to the container, the developer can rest assured that the application will run on any other Linux machine regardless of any customized settings that machine might have that could differ from the machine used for writing and testing the code.[1] Docker Here are some of the benefits of using Docker: Return on Investment and Cost Savings by dramatically reducing infrastructure resources [2]. Docker containers ensure consistency across multiple developments and release cycles, standardizing your environment [2]. Rapid Deployment since Docker creates a container for every process and does not boot an OS [2]. All major cloud computing providers support docker, including Heroku. Docker makes sure each container has its own resources that are isolated from other containers [2]. If you would like to take a deeper dive into docker, . I recommend this blog . Next: Install docker — for macOS, windows or Ubuntu Once Docker has been install run the following to make sure it is working properly. > docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES Heroku What is Heroku? Why Heroku? Heroku is a platform as a service (PaaS) that enables developers to build, run, and operate applications entirely in the cloud. I am going to use Heroku because it’s free to use for development and testing, easy to deploy (does not require much work), supports many languages by default, (As I could not find many resources on deploying TensorFlow models to Heroku, please leave comments below if you have trouble and I will help troubleshoot) Signup & Install heroku-cli Signup to Heroku Install heroku-cli After you have signed up to install heroku-cil, run the following commands to confirm the installation. > heroku --versionheroku/7.18.3 darwin-x64 node-v10.12.0

# make sure you have logged in to your heroku account> heroku login# Output should have:   Logged in as xxxxx@xxx.xx Nice work — now we are going to deploy TensorFlow Serving docker image to Heroku; however, since the default TensorFlow Serving docker image is not optimized to be deployed to Heroku, I have created a Dockerfile: Run the following code to build the docker image, making sure you are in the same folder as Dockerfile live. > docker build -t tf-serving-heroku-1.11 . Let’s run it locally to test it > docker run -p 8501:8501 -e PORT=8501 -t tf-serving-heroku-1.11...

2018-10-27 21:17:47.515120: I tensorflow_serving/model_servers/server.cc:301] Exporting HTTP/REST API at:localhost:8501 ... Call the model Now we are ready to call the production-ready model. Before calling the model, let’s understand how TensorFlow Serving RESTful API actually works: POST http://host:port/<URI>:<VERB>

URI: /v1/models/${MODEL_NAME}[/versions/${MODEL_VERSION}]VERB: classify|regress|predict In our case the request will look like: http://localhost:8501/v1/models/tf_serving_keras_mobilenetv2/versions/1:predict Second, the JSON data that is sent to TensorFlow Model Server has to be structured in a very particular way {  // (Optional) Serving signature to use.  // If unspecified default serving signature is used.  "signature_name": <string>,

// Input Tensors in row ("instances") or columnar ("inputs") format.  // A request can have either of them but NOT both.  "instances": <value>|<(nested)list>|<list-of-objects>  "inputs": <value>|<(nested)list>|<object>} In our case: {  "signature_name":'prediction',  "instances": [{"images":IMAGE_AS_LIST}]} You can see the full documentation about RESTful API . Here Now, I will show an example how to call our model from python Image before preprocess: Image size 1000 X 1000 Image after preprocess: Image size 96 X 96 It will print T-shirt/top If you have reached this far , you have overcome many hurdles, congratulations! — we are almost there. Now we are going to deploy the model to Heroku. Before running any of the commands below make sure you are in the same folder where Dockerfile live. Log in to Container Registry: > heroku container:login Create a Heroku app: > heroku create ${YOUR_APP_NAME} Push the docker image to Heroku: > heroku  container:push web -a ${YOUR_APP_NAME} To test your app, change the RESTful API URL to: { }/v1/models/ /versions/1:predict YOUR_APP_NAME tf_serving_keras_mobilenetv2 That is - you have now trained and deployed a deep learning model to production. Welcome to the very small production ready machine learning community :) I have created a , which has all the code for part 1 and 2 of this tutorial. Jupyter notebook Conclusion I hope you have enjoyed this 2 part series on deploying and training a deep learning model. I will be responding to question/suggestions, for whomever leaves them below in the comments section. If you would like to see more tutorials like this please follow, clap and suggest what we should learn next :) A special thanks to for his help and feedback. Shaun Gamboa References What is Docker? Top 10 Benefits of Docker Serving a TensorFlow Model

Training and Deploying A Deep Learning Model in Keras MobileNet V2 and Heroku: A Step-by-Step…

Part two of a two-part series: It’s like hot dog not hot dog, but for roads In part 1, , we covered how to retrain a on a new dataset. Specifically, we trained a classifier to detect Road or Not Road at more than 400 frames per second on a laptop. Creating Insanely Fast Image Classifiers with MobileNet in TensorFlow MobileNet MobileNets are made for — wait for it — mobile devices. So, let’s move our road not road model to an app so we can see it in action. Android Goals and Plan Let’s set some constraints so we have something specific to shoot for. We’ll attempt to: Retrain a MobileNet on a very small amount of purpose-built data Achieve 95% classification accuracy on a hold out test set Use less than 5% of a $300 device’s CPU while running inference To do that, we’ll follow these steps: Generate a new training dataset Train several MobileNet configurations to find the smallest net that will hit our accuracy target Get benchmarks by running Inception V3 on Android Update the TensorFlow Android example app to use our MobileNet model Try it in the wild Tune it to get below 5% CPU usage Building the Dataset In the previous post, we were classifying road/not road , so we pulled images from several sources. Now we’re going to drill down on the problem a bit more. If you recall, the purpose for this project is for user privacy: At , we’re building driving safety features for mobile devices that make use of the camera. So when someone turns on the app, we want to validate that what we’re looking at is a road. If we aren’t, we’ll disable recording. generally Coastline So to build our training set, I’m going to walk around doing everyday things while recording video: Around my house, outside my car, inside my car fiddling with the radio, petting the cats, etc. This will be our “not road” training data. Some “not road” examples. For our “road” data, I’m going to sample randomly from the Coastline driving dataset, which is from a camera facing out the front of a car. “Road” examples. Notice each image has a mount in view. We need to be careful that the network doesn’t simply learn to recognize that object. We’ll take care of that through data augmentation. With 3,000 examples from each set, we’re ready to train. Training MobileNet on our custom dataset The next step is to see what sort of accuracy we can get from the different MobileNet configurations. We’ll start with training the widest one: MoileNet 1.0 @ 128. And because we’re going to put this on a mobile device, we’ll use quantized weights, which will reduce the model memory footprint even further. For the details on how to retrain MobileNet on your own data, be sure to check out my previous post . From the root TensorFlow folder, we’ll run: python tensorflow/examples/image_retraining/retrain.py \  --image_dir ~/ml/blogs/road-not-road/data/ \  --learning_rate= After 1,000 training steps, we achieve 99.7% accuracy on our hold out set. Wow! Apparently MobileNets are pretty good at classifying roads. Here are a couple images it miss-classified: : A “not road” image classified as road. I’d say that’s an acceptable failure. It’s clearly a road, just not the type of road we’re looking for. : A “road” image classified as “not road”. I think this is because there were no bridges in the training set. Could be fixed with more data. Left Right Now let’s do the same thing, but with the smallest MobileNet: 0.25 @ 128, quantized. After 1,000 training steps, we get to 92.6%. Doesn’t satisfy our accuracy target. How about something a little wider, say 0.5? 95.0%! And the final model is just 1.6 MB. Looks like our money shot, if just barely. ( It should be noted that this entire model is trained on just 10 minutes of video captured at 10fps. There is a lot of room for improvement by piling on the data.) Let’s give it a quick try to make sure it’s working as expected: python tensorflow/examples/label_image/label_image.py \--graph=/tmp/output_graph.pb \--labels=/tmp/output_labels.txt \--image=/home/harvitronix/ml/blogs/road-not-road/test-image.jpg \--input_layer=input \--output_layer=final_result \--input_mean=128 \--input_std=128 \--input_width=128 \--input_height=128 Road: 0.99023 confidence. Looks good! And since our headline has the words “insanely fast” in it, how fast can we run this on an NVIDIA GeForce 960m GPU on our laptop? It runs through 1,000 images in just 3.36 seconds. That’s 297.6 frames per second! Using our MobileNet model in an Android app Now that we have a model that’s tiny, fast and accurate enough for our use case, let’s load it up in an Android app so we can test it in the real world. Don’t have a model trained yet? Download the model I trained on the data described above . It contains both the .pb and label files. Extract and follow the instructions below. Sticking with our theme of using tools provided by TensorFlow (how awesome is that, btw?), we’ll make use of the Android example project to get this thing running in no time flat. Getting and building the project If you haven’t already, go ahead and clone the TensorFlow repo: git clone https://github.com/tensorflow/tensorflow.git --depth 1 You’ll find an Android project ready-made for doing this kind of task in the folder. Open the folder in Android Studio, build it, load the APK on your phone, and you’ve got an image classifier that uses the Inception V3 model trained on ImageNet, ready to tell apart your cat from a platypus. tesnorflow/examples/android If you have trouble building the app, be sure to take a look at the instructions in the TensorFlow Android ReadMe . My biggest challenge was the NDK version. Downgrading to r12b solved my problems. Inception speed benchmark Since we’ve already got Inception running on the app, let’s take some measurements so we can compare it against our MobileNet. The Inception model that comes with the example project is 53.9 mb. That’s versus our MobileNet that’s just 1.6 mb! It runs at about 240ms per inference, or 4 frames per second (fps). And it uses about 40% of the CPU. Inception V3 running at 4fps. Let’s try it at 1fps: Inception V3 running at 1fps. Still up over 35%. Let’s hope our MobileNet can do better than that, or we’re not going to get anywhere near our goal of max 5% usage. Switching to MobileNet Now let’s make a couple minor changes to the Android project to use our custom MobileNet model. First, copy your model and labels into the project’s folder. Mine were at and . assets /tmp/output_graph.pb /tmp/output_labels.txt Next, open up ClassifierActivity, which can be found in: tensorflow/examples/android/src/org/tensorflow/demo/ClassifierActivity.java You’ll want to update the constants at the top of the file to define the settings for our new model. It looks like this when you first open it: private static final int = 224;private static final int = 117;private static final float = 1;private static final String = "input";private static final String = "output"; INPUT_SIZE IMAGE_MEAN IMAGE_STD INPUT_NAME OUTPUT_NAME private static final String = "file:///android_asset/tensorflow_inception_graph.pb";private static final String ="file:///android_asset/imagenet_comp_graph_label_strings.txt"; MODEL_FILE LABEL_FILE Change it to: private static final int = 128;private static final int = 128;private static final float = 128;private static final String = "input";private static final String = "final_result"; INPUT_SIZE IMAGE_MEAN IMAGE_STD INPUT_NAME OUTPUT_NAME private static final String = "file:///android_asset/output_graph.pb";private static final String ="file:///android_asset/output_labels.txt"; MODEL_FILE LABEL_FILE Hit run to build the project and load the APK on your device, and you’ve got your very own road / not road classifier! The Results Here’s a video of my road / not road app in action. I made a few tweaks to the UI to make it easier to see what’s going on: So how fast is it, and what about CPU usage? On my Xiaomi Mi5, this thing flies. It does inference in about 55ms, which is 18 frames per second! On a ! $300 Xiaomi Mi5s It is quite CPU intensive, though, using 25–30% when running at full-throttle. Makes sense, since we’re running it as fast as it will go. MobileNEt CPU monitor, running at 18fps. We want to get down to under 5%, which we’ll do by reducing the frequency at which it runs, since our use case doesn’t need to run inference continuously to achieve our privacy objective discussed above. Changing it to classify once every 18 frames (so once a second, roughly) brings the average usage down to about 5.5%! MobileNet running at 1fps. So our MobileNet model is 1/30th as large as Inception. It runs more than 3x faster per frame. And it uses far fewer CPU resources. It’s safe to say me and MobileNets are going to be like like: Enjoyed this post? Help others find it by hitting the little ❤. Thanks for reading!

Building an insanely fast image classifier on Android with MobileNets in TensorFlow

“It’s like hot dog not hot dog, but for roads.” MobileNets are a new family of convolutional neural networks that are set to blow your mind, and today we’re going to train one on a custom dataset. There are a few things that make MobileNets awesome: They’re insanely small They’re insanely fast They’re remarkably accurate They’re easy to tune for resources vs. accuracy Why is this important? Many mobile deep tasks are actually performed in the cloud. When you want to classify an image, that image is sent to a web service, it’s classified on a remote server, and the result is sent back to your phone. learning That’s changing quickly. The computational power on your phone is increasing rapidly, and the complexity required for computer vision is shrinking (thanks to architectures like SqueezeNet and MobileNet). network Besides the clear benefit of AI without an internet connection, sending images to the cloud is impractical in speed-hungry situations, like in the vehicle safety apps we’re developing at . Coastline With that, let’s learn the following: What are MobileNets? How to build a custom dataset to train a MobileNet with TensorFlow How to train a MobileNet that’s pretrained on ImageNet with TensorFlow How MobileNets perform against Inception V3 How to use your retrained MobileNet to classify images What are MobileNets? MobileNets are a class of convolutional neural network designed by researches at Google. They are coined “mobile-first” in that they’re architected from the ground up to be resource-friendly and run quickly, right on your phone. The main difference between the MobileNet architecture and a “traditional” CNN’s is instead of a single 3x3 convolution layer followed by batch norm and ReLU, MobileNets split the convolution into a 3x3 depthwise conv and a 1x1 pointwise conv. The details of why this is so significant can be found in the , which I strongly encourage you to read. MobileNet paper So what’s the catch? Accuracy. MobileNets are not usually as accurate as the bigger, more resource-intensive networks we’ve come to know and love. But finding that resource/accuracy trade-off is where MobileNets really shine. MobileNets surface two parameters that we can tune to fit the resource/accuracy trade-off of our exact problem: width multiplier and resolution multiplier. The width multiplier allows us to thin the network, while the resolution multiplier changes the input dimensions of the image, reducing the internal representation at every layer. Google and released 16 ImageNet checkpoints, each corresponding to a different parameter configuration. This gives us an excellent starting point for training our own classifiers that are small and fast. open-sourced the MobileNet architecture insanely insanely To learn more about how MobileNets work, read MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications . How to build your own dataset to train a MobileNet Our challenge today is to build an image classifier that can detect if an image is a road or not a road. It’s like , but for roads. hot dog, not hot dog Why road, not road? At , we’re developing safety features for your car with mobile apps that use computer vision. As with all vision problems, user privacy is critical. So one of the first checks we do when a user turns on the camera in our app is check if it sees a road. If it doesn’t, we turn off the camera. We want to be able to do this fast and at as little computational cost to the user as possible. Coastline Since we’re tackling a custom problem, we need to start with creating our dataset. Our target is to collect 10,000 images, split roughly evenly 50/50 road/not road. We’ll get data from a few different places: 4,000 obviously road images sampled randomly from the Coastline driving dataset 2,000 obviously not road images sampled randomly from the ImageNet dataset 3,000 less obviously not road scenics sampled from the internet to make sure the classifier doesn’t just learn sky, not sky 1,000 less obviously road scenics also sampled from the internet to make sure the classifier doesn’t recognize windshield reflections or other commonalities in the Coastline data We’ll place each image into one of two folders, each representing the class of that image: and . That’s all we have to do to prepare our images for retraining! road not-road However, while grabbing from the internet is a great place to add diversity to your dataset, it comes with a drawback: the labels are noisy. For instance, an image found by searching “road landscape” could have a road front and center with a nice scene in the background, or it could be a mountain scene with a tiny road off in the distance. CC images Two images both labeled “road”. For our problem, the first should be “not road” and the second should be “road”. Left: by is licensed under Right: by is licensed under “Scenic Mountain Road” Tyler Thompson CC BY 2.0 “150927-road-scenic-bypass-mountains.jpg” r. nial bradshaw CC BY 2.0 We could solve for this by going through each image and hand-labeling it, but why do that when we have deep learning?! Instead, we retrain a big network (like Inception V3) on all our data, paying special attention not to over-fit on our training data by early stopping and heavy data augmentation. Then we run every image of our dataset (even those images we just used to train!) through the network and keep track of the images it classified incorrectly or with little confidence. Then we go through each of those images and move them to their proper classes, if applicable. This reduces the number of images we have to manually clean up significantly. And doing multiple passes of this technique helped us increase our accuracy by seven percentage points on Inception. Now that we have 5,000 road images and 5,000 not road images, in a structure like this… data/road/[images...]not-road/[images...] …we’ll use TensorFlow and transfer learning to fine-tune MobileNets on our custom dataset. How to retrain a MobileNet that’s pretrained on ImageNet TensorFlow comes packaged with great tools that you can use to retrain MobileNets without having to actually write any code. This stuff is fresh off the presses: Retraining support for MobileNet was ! added less than a week ago A couple very obviously road images from the Coastline dataset. If you don’t have it downloaded already, fork and/or clone the TensorFlow repo: git clone https://github.com/tensorflow/tensorflow.git Now you can use the scripts in the example folder to retrain MobileNet on your own data. But wait! Which MobileNet should you use? That’s a good question. Let’s retrain a small assortment and see how they perform. To kick off training, we’ll run the following command from the root of the TensorFlow repo: python tensorflow/examples/image_retraining/retrain.py \    --image_dir ~/ml/blogs/road-not-road/data/ \    --learning_rate=0.0001 \    --testing_percentage=20 \    --validation_percentage=20 \    --train_batch_size=32 \    --validation_batch_size=-1 \    --flip_left_right True \    --random_scale=30 \    --random_brightness=30 \    --eval_step_interval=100 \    --how_many_training_steps=600 \    --architecture mobilenet_1.0_224 The flag is where we tell the retraining script which version of MobileNet we want to use. The 1.0 corresponds to the width multiplier, and can be 1.0, 0.75, 0.50 or 0.25. The 224 corresponds to image resolution, and can be 224, 192, 160 or 128. For example, to train the smallest version, you’d use . architecture --architecture mobilenet_0.25_128 Some other important parameters: learning_rate: This is something you’ll want to play with. I found 0.0001 to work well. testing and validation percentage: The script will split your data into train/val/test for you. It will use train to train, val to give performance updates every “eval_step_interval”, and test will run after “how_many_training_steps” to give you your final score. validation_batch_size: Setting this to -1 tells the script to use all your data to validate on. When you don’t have a lot of data (like only 10,000 images), it’s a good idea to use -1 here to reduce variance between evaluation steps. After retraining on several model architectures, let’s see how they compare. Comparing MobileNet parameters and their performance against Inception After just 600 steps on training Inception to get a baseline (by setting the flag to , we hit 95.9%. Training took 18 minutes. (There is a lot of room for improvement here, but we don’t have all day!) The resulting checkpoint landed at 84mb. And doing a quick speed test by running 1,000 images through it shows it can classify images on an NVIDIA GeForce 960m GPU at ~19fps. — architecture inception_v3) Aside: Why “only” 95.9% and not 100%? It seems like a pretty simple problem, right? Well, besides the ample tuning we could do to the training parameters (we actually achieved 98.9% with the same data using a different configuration in another go), it turns out the distinction between classes is a bit more subtle than it seems on the surface. Take these cases: The image is a one-lane dirt road in the woods: Is it a road or a trail? I don’t even know. It’s a landscape scenic with a road in the distance. Is it a picture or a road, or is there just a road in the picture? At what point does the scenic move classes? It’s an artsy tilt-shift photo of a couple holding hands in the foreground and a time-lapsed street in the background. Road or not road? Flip a coin. So, how do the MobileNets perform? Not surprisingly, not quite as well. However, the tradeoff benefit is . astounding Using the biggest MobileNet (1.0, 224), we were able to achieve 95.5% accuracy with just 4 minutes of training. The resulting model size was just 17mb, and it can run on the same GPU at ~135fps. For those keeping score, that’s . All for just 0.4% loss in accuracy. 7 times faster and a quarter the size How about the smallest MobileNet (0.24, 128), using quantized weights? Big accuracy tradeoff, achieving just 89.2%. But get this: 450 frames per second, and the model takes just 930kb of memory. That’s kilobytes! Using your retrained MobileNet to classify images Now that you have your MobileNet retrained on your custom dataset, it’s time to give it a try. Not surprisingly, TensorFlow comes with a script to do that, too. python3 tensorflow/examples/label_image/label_image.py \--graph=/tmp/mobilenet_0.50_192.pb \--labels=/tmp/output_labels.txt \--image=/home/harvitronix/ml/blogs/road-not-road/test-image.jpg \--input_layer=input \--output_layer=final_result \--input_mean=128 \--input_std=128 \--input_width=192 \--input_height=192 Our network classifies this as road, with confidence of 0.686811. Not too confidence, but hey, that’s a FAST road! Aside: It should be noted that in our fairly simple two-class problem, the accuracy trade-off is not that big. In the case of ImageNet with its 1,001 classes, the accuracy tradeoffs are much more significant. See the table here . Next steps Okay, so the whole point of MobileNets is to run on mobile, right? Stay tuned! In our next post, we’ll create some purpose-built training data, fine-tune again, and load our retrained MobileNet into an Android app. We’ll see how fast it can run on a mobile device and how accurate it is in the real world. : Part 2 is now live! Don’t miss . UPDATE Building an insanely fast image classifier on Android with MobileNets in TensorFlow

Stories, News & Mentions about MOBILEGPT

Mobilegpt’s Stories on HackerNoon

Mobilegpt's News & Mentions