Oatfi’s APIs allow B2B payment platforms to launch products such as ‘BNPL’, factoring, receivables financing, and other short term credit solutions where users have working capital needs

OatFi

An Irate Neighbor

How the Alibaba tech team uses deep learning to help online marketplaces drive strong business results through the creation of quality outfit displays Maintaining success in online retail is a constant struggle that requires retailers to determine the best method of presenting their products to customers. One method is to make an attractive poster for the item, which previously required a great deal of human labor. Another way is suggesting item combinations — especially for clothes and accessories — in a bid to encourage customers to make an extra purchase. Alibaba’s Dida AI assistant combines both of these strategies. Dida is modeled after Alibaba’s famous poster designing AI engine, Lu Ban. Like Lu Ban, Dida uses graphical algorithms to collate data and train itself to generate targeted, high-quality images of popular item combinations. While mainly used for generating personalized outfit combinations for online fashion retail, Dida’s use has now expanded to other areas, such as copywriting for product description. So, how does Dida work? Technically Dida is not a single entity but a combination of multiple platforms and algorithms, including front-end operations, algorithm, image mosaic, and personalized launch platforms. First, operators select items on the front end and designate items to be combined using deep image processing and combination algorithms. When one or several items trigger a combination request to create a combination, Dida searches for additional items that are a good match and follow specific operating rules. Next, descriptive titles are generated based on information from the triggering and returned items. Finally, the items are synthesized and displayed in a visually flattering way using smart typography. The images of the item combinations are then generated, personalized, and pushed to users with recommendation algorithms. As of February 2017, Dida’s algorithms have been launched across a variety of Alibaba’s eCommerce platforms, including Taobao and Tmall. For major promotional events, Dida aids operators in generating millions of combinations for a wide range of products and it is utilized by hundreds of thousands of Alibaba design experts. Alibaba is already recognizing positive results from trials on several of its pages, including iFashion, Mobile Taobao Primary Focus, and YouHaoHuo. Images from a backend outfit display on Dida The next section briefly summarizes the top three benefits of using the Dida platform, and the rest of this article discusses the technical foundations of the Dida platform and the success stories of iFashion, Mobile Taobao Primary Focus, and YouHaoHuo. Dida’s Benefits in a Nutshell 1. Content production Dida utilizes a deep learning network to collate extensive collections of information from users, products, and operations knowledge. When input to help with content creation, the Dida AI engine performs at the same level of quality as an experienced design expert. The graphical algorithms generated by Dida are then applied across multiple business areas. 2. Platform enabling Dida combines graphical algorithms and platform establishment to allow operators to select items from the Dida platform and generate item combinations and personalized launches in a one-stop management process. 3. Improved efficiency The average design expert produces thousands of item combinations daily. The Dida platform generates millions of item combinations per hour, which is a massive improvement in efficiency. Dida also combines algorithms to expand the overall size of the information pool and enhance personalization. Technical Foundations of Dida When used for online fashion retail, Dida generates outfit combinations and descriptions, which requires designing graphical and textual algorithm frames. All bottom layer data is shared, including pictures and titles for items, operational input data, and various other information. Dida utilizes this information to generate outfits and then generates descriptions for each one. First, a picture combination algorithm uses Convolutional Neural Networks (CNN) to pre-process images. Next, two combination logic algorithms are run using the Deep Semantic Similarity Model (DSSM): - Long Short-Term Memory (LSTM)-based sequencing outfit production. - Deep Aggregated Network (DAN)-based non-sequencing outfit production. The generated outfits are imported into Context-aware Pointer-Generator Networks (CPGN) to generate text descriptions for each outfit. The final result includes text and graphs that provide holistic descriptions of each combination. The relevant image and textual algorithms are introduced in the following sections. Image Algorithms This section includes all relevant preparation and processes for the use of image algorithms in this scenario. Data Alibaba’s training data was initially sourced from Polyvore, a website that stores massive amounts of user-submitted item combination examples that are further delineated based on factors such as user likes and comments. When Taobao’s internal businesses practices were firmly established, Alibaba created an in-house database using hundreds of thousands of high-quality outfit combinations obtained from veteran Taobao experts. Characterization First, items must be characterized. Images are the most intuitive in terms of delivering information. Alibaba uses Lu Ban’s library of millions of white-background images to generate image characteristics for items in a specific item pool. This is accomplished using the inception v3 model of CNN. The process is as follows: 1. The pre-trained model is fine-tuned using the category as a label, and the vector expression of the penultimate layer is extracted and used as the image representation of the item. 2. Vector expressions for all images with category constraints are organized into clusters using K-means clustering. Some optimizations have been done on K-means clustering which ensure centralized and balanced distribution, making it ideal for use in this scenario. Optimized K-means clustering uses differences in category relationships and combinations and the distributed quantity of items for various categories as determiners. An individual category can include a variety of different clusters. This step ensures every item ends up in a cluster. 3. The model is fine-tuned again using inception v3, with the clusters generated in Step 2 being used as labels. High-dimension vector expressions are also extracted. These vector expressions are used as the final characteristics for images. Model I: LSTM-based sequencing outfit combination production 1. High-dimension vectors are obtained using CNN, and the vectors of side information are embedding- and stacking-transformed as the input layer of the model. 2. Input vectors are divided into two routes after the first MLP layer. One route leads to the LSTM network for sequence learning, and the other leads to the DSSM network for vector alignment. The details of both are below. LSTM network Generating combinations is a sequential process. Each item generated for an outfit is considered a sequential step. Starting from the first item, each new item must be correlated with previously-generated items. This is possible due to the LSTM network’s intrinsic sequential relationships. LSTM, an extension of Recurrent Neural Network (RNN), includes function gates that allow it to capture long-term dependency effectively. For this scenario, S represents an outfit, xt represents the CNN characteristic representation of the item, and S=x1,x2,…,xN represents an outfit sequence. According to the maximum likelihood estimate (MLE), the main aim was to maximize the expectation of: DSSM network A DSSM network approach is adopted due to the expectation that items in one outfit combination should have closer distances in vector space. Positive samples come from online logs and quality outfits. The most-clicked and highest-quality outfits from veteran Alibaba experts are collected from the online log and divided into pairs to serve as positive examples for the DSSM network. Outfits receiving a low number of clicks are used as negative examples. As shown in the figure, vector expressions for items pass through MLP before entering LSTM. When LSTM generates an item X, X is subjected to MLP transformation, and its distance from other items that have not yet entered LSTM is calculated. Positive examples of other items are denoted as Y+ and negative examples as Y−. Short distances between positive examples and long distances between negative ones are the ideal situation. Therefore, loss is expressed as: where the sim function uses cosine similarity, θ stands for parameters, and the goal is to maximize Δ. Mini-batch SGD is used to optimize θ on GPU. Model II: DAN-based non-sequencing outfit production The launch of the first version based on the LSTM Model proved satisfactory. However, subsequent studies indicated DAN as a superior model. The only changes made to the previously described structure is the replacement of the LSTM module. The DAN model achieved lower loss and better output. The core function of DAN is considering outfits as a combination mode rather than a sequencing mode. Take outfits consisting of tops and pants as an example. Two sequences of training data are required to train the LSTM network: Tops + pants and pants + tops. For DAN, tops and pants are input into the network as combinations without sequential differences. An outfit’s training data is input into DAN after passing through CNN and side-information embedding. Vectors are changed to a non-linear layer first, and then they enter the pooling layer. Trials of this process using sum-pooling and max-pooling have found the former is the better option. The entire DAN-based process is outlined in the following figure. Compared to LSTM, DAN achieves lower loss during training. Another advantage of DAN is that it only requires combination data, rather than full permutations, to construct training data. Less training data shortens training time and makes a periodical iteration model possible. Context graph-based prediction process Outfits tend to be defined differently based on the operator. Women’s clothing sellers often view top+bottom+shoes or dress+accessories+bag as complete outfits. One-piece dresses and jeans never appear in the same outfit. From the perspective of a seller of household goods, the absence of either a bed, a nightstand, lamps, or wallpaper can make a bedroom combination incomplete. In practice, operators often factor in the usage scenario and additional restrictions based on style and season. The main difficulty in this scenario is creating combination algorithms that meet the needs of operators. Alibaba designed a context graph to address this issue. A context graph is a set of structured operation rules, including constraints on category, style, season, and more. In the outfit combination prediction stage, all items and their side-information undergo embedding, stacking and full connections before they are stored in an item pool. Take DAN as an example: when an activity sends requests, all triggered items undergo DAN. Without considering constraints, the MLP output is used to search the item pool for similar items to obtain the next item. According to the context graph, when constraints are factored into the similarity search, results are filtered so that only items that meet operation rules end up in the candidate set. Next, TopK is picked out of the candidate set. Whenever an outfit is generated, all satisfied combination constraints are re-calculated using the algorithms to promote the generation of the next item and a new candidate set. The context graph is packed into the model, making outfit combination prediction a fully real-time process. It also ensures a high yield rate, as every generated outfit meets the input conditions of operators, reducing the cost of manual filtering. Textual algorithms This section describes all relevant preparation and processes for the use of textual algorithms in this scenario Data Veteran design experts have created hundreds of thousands of self-determined outfits. Descriptions of these outfits are collated and used as training data. Training data inputs include item titles and outfit tags. Words are used as the basic unit. CPGN model Operational inputs are added based on PGN to establish strong connections between copywriting, items, and operational requirements. This new approach is named CPGN, and its algorithm structure is as follows: The entire frame comprises an encoder-decoder structure. First, the original data (x1, x2, …, xn) and operational inputs (z1, z2, …, zn) are encoded. Every individual word from the original data goes through a single layer of a bi-directional LSTM network, and the hidden state is denoted as hi. Operational input can be a complete sentence or keywords. In the former case, LSTM processing is still adopted. In the latter case, embedding is applied directly to keywords, which are denoted as ri. In scenarios where i represents the i-th input, and t represents the number of decoding steps, the attention distribution (at) and context vectors (ht∗ and rt∗) are denoted as follows: where η is a multi-layer MLP with tanh being its activation function, and st represents the decoder status at the t-th step. The attention distribution is viewed as the probability for each encoded source word to generate the decode, where ht∗ and rt∗ are weighted sums of attention and the expression of the information obtained from source statements. The probability distribution of the next word across the dictionary can be derived on this basis: where g is a two-layer MLP. In this way, the probability of the generated section is derived. To balance pointing and the generator, parameter pgen∈[0,1] is designed, which is a probability soft-switch associated with the current decoder status st, context vectors ht∗, rt∗, and decoder input yt−1. pgen allows the next word to be generated from the dictionary or copied from the input. Assuming the predicted probability of each word in the dictionary is Pvocab(w), then: where σ is a sigmoid function. The current P(w) contains the entire dictionary, including words that were not previously in the dictionary but appeared in inputs. This helps to solve problems with OOV. In the training stage, assuming the t-th target word is wt∗, loss can be denoted as: Finally, a coverage algorithm was introduced to tackle repeat words. As all algorithms are launched directly online, Alibaba added a regular expression layer following duplicate removal in the algorithm to eliminate loopholes. Dida Case Studies This section looks at some successful case studies of Dida in iFashion outfits, Mobile Taobao Primary Focus, and YouHaoHuo. iFashion outfit iFashion is a Taobao page that uses an outfit combination-centric scenario. iFashion maintains high-performance requirements for graphical algorithms, both in terms of content quality and visual effect. Alibaba periodically uses Dida to produce matching outfits and text descriptions for iFashion selections. This represents a huge addition to the outfit pool, which previously relied solely on contributions from veteran experts. Outfits generated via algorithm are mixed together with outfits from veteran experts in waterfalls and then displayed to users in personalized pushes. Creating outfits via algorithm provides significant benefits, including decreased costs, positive feedback, and high conversion rates. Mobile Taobao Primary Focus Mobile Taobao Primary Focus is a Taobao banner with strong operational demands. Every image is linked to an event page that displays copywriting for the event and images of the products being sold. In personalized launches, Alibaba’s recommended algorithms displayed personalized contents for every user based on their behavior. Before Dida, identical copywriting was used for all item combinations, even if the images were personalized. In this scenario, Dida’s graphical algorithms were used to create multi-item combinations for the garment industry. The resulting outfit combinations were high-quality and enriched the imagery and presentation in the primary focus area. Alibaba has also tested using textual algorithms for copywriting across multiple industries. When using a conventional approach, operators often use a very bland, general style. For example: · Primary copywriting: Sports party · Secondary copywriting: Choice brands, large discount · Benefits: Rush for big value coupons. This style of copywriting is boring and easily ignored by consumers. In contrast, Dida generates copywriting that is far more interesting and engaging: For this example, Alibaba’s model uses item titles, descriptions, and attributes (information about yoga bricks) as the input data. Operational inputs are individual benefit keywords or tags (quality choices, big brands, sales promotions). The final copywriting consists of: · Primary copywriting: Quality choices for yoga bricks (a combination of words extracted from item titles and operational inputs) · Secondary copywriting: Big brands, big discounts, spend-and-get-back (derived from words extracted from operational inputs). In comparison, smart copywriting is customized for specific items and events and describes products in a way that is interesting and highlights their benefits and associated sales events. The graphs and text combinations generated for Primary Focus resulted in double-digit percentage increases in CTR and UCTR. YouHaoHuo(有好货) YouHaoHuo is a Taobao page that is highly popular among consumers due to its flagship concept, “Haohuo,” which refers to the sale of quality items. Many of the YouHaoHuo product titles designed by veteran experts are too long to be fully displayed on the current page layout. The resulting truncation leads to unclear item descriptions and prevents users from viewing the full information for items. Alibaba used Dida copywriting algorithms to re-extract key information from titles and limited the length, helping users to make better decisions. (Original article by Chen Wen陈雯) Alibaba Tech First hand and in-depth information about Alibaba’s latest technology → Facebook: . Twitter: . “Alibaba Tech” “AlibabaTech”

Finding the Perfect Outfit with Alibaba’s Dida AI Assistant

Juxtapositions in Making a Home I’m walking down poorly-lit Bancroft on a late-evening stroll. Then, I pass by Urban Outfitters — it is 11pm, three hours after closing, yet the high-ceilinged lights are all lit up, illuminating the vast interior of cheaply-made and overpriced clothing. On the storefront patio is a camp of young, long-haired, men. They have made their place of rest for the night — duffle bags, overstuffed, and sleeping bags spread flat on the concrete floor. Behind them is a store-box of light, showing off the window-display mannequins and window decal. One of them advertises: UO Home. homeless In her special, Ali Wong — a San Francisco native herself — shared a frustration: “That store Urban Outfitters has made things very confusing for my generation.” She joked, “You homeless or you a hipster? Is that beard for fashion or for warmth?” Baby Cobra Passing by, all I could do was chuckle and heartbreak at the irony. This is not my first time living in or the Bay Area. Yet, returning after a couple of years, the everyday juxtaposition of the haves versus have-nots, the Bay Area and Silicon Valley of my dreams versus the reality of displacement for so many, aches my heart. This is a place I love so much, where I want to call home and make a home — but, at what cost? Berkeley After my stroll, I spoke with my grandmother on the phone. “Is it safe?” She asked of the neighborhood where I walked. “Are there a lot of black people?” I deflected and said, “There’re all kinds and colors of people here.” The homeless aren’t just on the streets —they also live in cars, vans, RVs, and alternative, make-shift homes all over the Bay. Homeless, but making a home for themselves — or in some cases, for their families — on their own terms. Meanwhile, I’ve made a temporary home here at Berkeley, working on my thesis research, trying to understand how people pay attention to culture at tech companies. One of the many voices in this culture discourse is about authenticity and belonging — bringing your whole self to work and finding belonging in a company that cares (or at the very least, wants to, and tries to). That sounds like home to me. What does it mean for all these tech companies to be invested in making homes at work — homes of purpose and communities of people — when having an actual home to return to is a precarious reality for so many? A few nights later when I passed by Urban Outfitters, there was no one. Then, a few nights after that, I saw a woman curled up under her two blankets, fast asleep. She looked so tender I desperately wanted to take a picture, but didn’t. How much I wanted to turn off the lights and say goodnight, but couldn’t. Thank you so much for reading + sharing your thoughts. You can find more of me on Instagram , at my poetry blog , and at my website . Want my articles in your inbox? — subscribe here.

Home(less) at Urban Outfitters

If you've been following the VR industry, you know it's been around for decades. Companies have been trying to make virtual reality a mainstream success for ages. While Apple coined the buzz-worthy term “spatial computing” with their Vision Pro headset, they're certainly not the first to try. From the failed Virtual Boy in the 90s to the moderately successful Oculus and HTC Vive VR headsets, tech giants have struggled to find that elusive killer app. But if you just scroll through TikTok or Instagram, you'll see what's really on people's minds these days. And it definitely isn't about bigger screens to watch Netflix or play games. No, what's driving the conversation is beauty, looks, fashion, and how to glow up. So, as transformative as spatial computing could become, the multi-trillion-dollar global beauty and fashion space will lead AR. Our universal desire to look and feel good provides the real-life use case everyone’s looking for. The promises and challenges of AR beauty The core premise of AR beauty is simple but compelling: Using advanced computer vision and rendering to virtually overlay any makeup look, hairstyle, outfit, or accessory onto your face and body with hyper-realistic precision. Consumers can experiment endlessly, sampling different looks and products with no mess, effort, or commitment. If you've ever ordered a piece of clothing online only for it to look completely different (or just not fit right) when it arrives, you know how frustrating that can be—or bought an expensive watch or pair of kicks, only to regret the purchase when it doesn’t look quite how you expected. What if you could virtually try on anything before committing your money? That's the big idea of AR beauty and fashion—getting rid of the guesswork and potential disappointment by allowing you to realistically experiment with any look or product before pulling the trigger. But, some rudimentary AR makeup apps illustrate the significant technological challenges involved. Poor facial tracking, clunky interfaces, and unconvincing visual realism hamper the experience. Faithfully simulating the intricacies of how cosmetics interact with diverse facial geometries, skin tones, and movements at a high fidelity is an enormously complex computational problem. Implementing AI and AR technologies The latest breakthroughs in AI and AR technologies have proven useful in cracking these challenges. For example, Perfect Corp's AI Makeup Virtual Try-On uses machine learning to realistically render different makeup colors and textures like lipstick, eyeshadow, and more, with the simulations customized to each individual's unique facial features and microexpressions for an ultra-personalized virtual experience. If you're not all that into fashion and beauty, this may not seem like a big deal. But for those obsessed with looking their best, the ability to visualize any look or product before buying could be hugely transformative. Their AI Skin Analysis tech goes further, using artificial intelligence to scan faces and detect 14 different skin conditions and concerns, then surfacing customized product and skincare recommendations tailored to the user. And the AgileHand technology allows realistic virtual try-ons to extend beyond just faces by precisely tracking hand motions. This enables photorealistic AR visualizations for intricate products like jewelry, watches, and nail polish. Some other AR beauty solutions include Makeup AR for virtual makeup try-ons, AI Foundation Shade Finder and Matcher for accurate skin tone detection and foundation shade matching, AI Skin Analysis for accurate skin concern detection and personalized product recommendations of products, and Emulation Technology for Skincare Treatment for accurate representations of the expected results of a course of treatment, overlaid directly onto their faces with the latest advancements in AR technology. The AI tech for AR beauty The recent renaissance of AI breakthroughs, especially around areas like generative adversarial networks (GANs), facial semantic segmentation, and neural rendering, is what's enabling the kind of hyper-realistic and personalized AR experiences that were merely SciFi just years ago. Using GANs, AI models can now learn to generate highly realistic representations of human faces, facial attributes, and facial animations from just a handful of source images. This opens up possibilities for instantly generating virtual makeup looks and hairstyles that are custom-tailored to each user's individual facial geometry and characteristics. Facial semantic segmentation allows AI to precisely understand and map out different facial features like skin, facial hair, eyelashes and more - enabling AR effects to seamlessly integrate and move in harmony with the user's face. Neural rendering techniques leveraging AI can also realistically synthesize the complex ways light interacts with different materials and textures in real time, far beyond what's possible with traditional rendering pipelines. Looking ahead, AI is set to transform AR beauty experiences into something much more akin to having a personal beauty guru. Using deep learning, an AI assistant could continuously analyze your unique attributes, habits, and preferences, automatically surfacing ideal products, trendy looks, and personalized recommendations for you to try virtually in AR with just a voice command or subtle gesture. The possibilities extend far beyond just virtual try-ons. AI could help formulate bespoke beauty products optimized for your individual needs and environmental conditions. Or DTC brands could let customers co-create new cosmetic products and shades with the assistance of AI. The takeaway: Beauty and fashion are the AR killer app While the metaverse is still largely vapourware, the humble makeup tutorial may push spatial computing into the mainstream sooner than we think. With an insatiable consumer appetite for looking our best and plenty of money to be made, the global beauty and fashion industry has both the motivation and resources to drive AR innovation and adoption harder than any other sector. If executed well and key technical barriers around rendering, tracking, and hardware are overcome, AR holds the transformative potential to reshape how we discover, purchase, use, and even perceive beauty products. Immersive virtual try-on experiences, AI-driven personalization, and participatory product development could not only reinvent the beauty consumer experience but also make individual self-expression more accessible than ever before. So, as the tech world covers each new OS update and app ecosystem announcement, keep your eye on the influencers and beauty fanatics pioneering the spatial computing frontier with a fresh face of makeup in every video. They may low-key catalyze the next ubiquitous computing paradigm while just trying to look fly. If you've been following the VR industry, you know it's been around for decades. Companies have been trying to make virtual reality a mainstream success for ages. While Apple coined the buzz-worthy term “ spatial computing ” with their Vision Pro headset, they're certainly not the first to try. spatial computing spatial computing From the failed Virtual Boy in the 90s to the moderately successful Oculus and HTC Vive VR headsets, tech giants have struggled to find that elusive killer app. But if you just scroll through TikTok or Instagram, you'll see what's really on people's minds these days. And it definitely isn't about bigger screens to watch Netflix or play games. No, what's driving the conversation is beauty, looks, fashion, and how to glow up. So, as transformative as spatial computing could become, the multi-trillion-dollar global beauty and fashion space will lead AR. Our universal desire to look and feel good provides the real-life use case everyone’s looking for. The promises and challenges of AR beauty The core premise of AR beauty is simple but compelling: Using advanced computer vision and rendering to virtually overlay any makeup look, hairstyle, outfit, or accessory onto your face and body with hyper-realistic precision. Consumers can experiment endlessly, sampling different looks and products with no mess, effort, or commitment. If you've ever ordered a piece of clothing online only for it to look completely different (or just not fit right) when it arrives, you know how frustrating that can be—or bought an expensive watch or pair of kicks, only to regret the purchase when it doesn’t look quite how you expected. frustrating frustrating What if you could virtually try on anything before committing your money? That's the big idea of AR beauty and fashion—getting rid of the guesswork and potential disappointment by allowing you to realistically experiment with any look or product before pulling the trigger. But, some rudimentary AR makeup apps illustrate the significant technological challenges involved. Poor facial tracking, clunky interfaces, and unconvincing visual realism hamper the experience. Faithfully simulating the intricacies of how cosmetics interact with diverse facial geometries, skin tones, and movements at a high fidelity is an enormously complex computational problem. Implementing AI and AR technologies The latest breakthroughs in AI and AR technologies have proven useful in cracking these challenges. For example, Perfect Corp's AI Makeup Virtual Try-On uses machine learning to realistically render different makeup colors and textures like lipstick, eyeshadow, and more, with the simulations customized to each individual's unique facial features and microexpressions for an ultra-personalized virtual experience. AI Makeup Virtual Try-On If you're not all that into fashion and beauty, this may not seem like a big deal. But for those obsessed with looking their best, the ability to visualize any look or product before buying could be hugely transformative. Their AI Skin Analysis tech goes further, using artificial intelligence to scan faces and detect 14 different skin conditions and concerns, then surfacing customized product and skincare recommendations tailored to the user. AI Skin Analysis And the AgileHand technology allows realistic virtual try-ons to extend beyond just faces by precisely tracking hand motions. This enables photorealistic AR visualizations for intricate products like jewelry, watches, and nail polish. Some other AR beauty solutions include Makeup AR for virtual makeup try-ons, AI Foundation Shade Finder and Matcher for accurate skin tone detection and foundation shade matching, AI Skin Analysis for accurate skin concern detection and personalized product recommendations of products, and Emulation Technology for Skincare Treatment for accurate representations of the expected results of a course of treatment, overlaid directly onto their faces with the latest advancements in AR technology. The AI tech for AR beauty The recent renaissance of AI breakthroughs, especially around areas like generative adversarial networks (GANs), facial semantic segmentation, and neural rendering, is what's enabling the kind of hyper-realistic and personalized AR experiences that were merely SciFi just years ago. Using GANs, AI models can now learn to generate highly realistic representations of human faces, facial attributes, and facial animations from just a handful of source images. This opens up possibilities for instantly generating virtual makeup looks and hairstyles that are custom-tailored to each user's individual facial geometry and characteristics. Facial semantic segmentation allows AI to precisely understand and map out different facial features like skin, facial hair, eyelashes and more - enabling AR effects to seamlessly integrate and move in harmony with the user's face. Neural rendering techniques leveraging AI can also realistically synthesize the complex ways light interacts with different materials and textures in real time, far beyond what's possible with traditional rendering pipelines. Looking ahead, AI is set to transform AR beauty experiences into something much more akin to having a personal beauty guru. Using deep learning, an AI assistant could continuously analyze your unique attributes, habits, and preferences, automatically surfacing ideal products, trendy looks, and personalized recommendations for you to try virtually in AR with just a voice command or subtle gesture. The possibilities extend far beyond just virtual try-ons. AI could help formulate bespoke beauty products optimized for your individual needs and environmental conditions. Or DTC brands could let customers co-create new cosmetic products and shades with the assistance of AI. The takeaway: Beauty and fashion are the AR killer app While the metaverse is still largely vapourware, the humble makeup tutorial may push spatial computing into the mainstream sooner than we think. With an insatiable consumer appetite for looking our best and plenty of money to be made, the global beauty and fashion industry has both the motivation and resources to drive AR innovation and adoption harder than any other sector. If executed well and key technical barriers around rendering, tracking, and hardware are overcome, AR holds the transformative potential to reshape how we discover, purchase, use, and even perceive beauty products. Immersive virtual try-on experiences, AI-driven personalization, and participatory product development could not only reinvent the beauty consumer experience but also make individual self-expression more accessible than ever before. So, as the tech world covers each new OS update and app ecosystem announcement, keep your eye on the influencers and beauty fanatics pioneering the spatial computing frontier with a fresh face of makeup in every video. They may low-key catalyze the next ubiquitous computing paradigm while just trying to look fly.

Why the Beauty and Fashion Industry is the Future of AR and Spatial Computing

Previously… https://awstip.com/reducing-eks-windows-node-5-min-start-time-to-90s-eks-windows-bootstrapper-4314f72367a0?embedable=true I threw in some tricks I used to make our Windows EKS nodes start in 90s rather than the 5 minutes a standard EKS Windows node takes. I introduced you to the git repo eks-windows-bootstrapper - the application replacing the AWS-provided configuration scripts for EKS Windows. https://github.com/atg-cloudops/eks-windows-bootstrapper?embedable=true In today’s story, I will walk you through how to make your Windows nodes take ~30s to start! Recap In the previous story, we got to the point where our start times looked like this: Let’s take a closer look at our latencies now: In order for us to get better launch times, we now have 2 areas where we can improve timing Windows Startup Configure Network/Start K8s Services Improving Windows Startup Performance As we are using the AWS Image Builder (https://aws.amazon.com/image-builder/) to build our custom AMIs as discussed in the previous story, any changes we make to our image need to be made here. Outside AWS Fast Launch - there really isn’t much we can do to help with the Windows boot process, if you are able - you can uninstall Windows Defender which will massively improve start performance. Side note - I found in our use case we do not actually need Fast Launch as we do not actually need to run the OOBE process. We can set our culture using Set-Culture, etc. inside one of our components, and as long as we remove the sections in Unattend.xml related to OOBE, Windows will skip much of the setup process that EC2 Fast Launch deals with. When you start the output AMI, it will start as quickly as a Fast Launch AMI, but without the added expense of pre-provisioning EBS snapshots. The downside to this approach is that you won’t be able to configure Windows settings on the first boot, what you bake into the AMI is what you will start with - So make sure you set culture/timezone information correctly somewhere in one of your Image Builder components. For reference: So with all avenues to make Windows do less stuff on boot exhausted, there was only one choice left… Changes were made to the EKS Windows Bootstrapper code so it could be run as a Windows Service. The idea behind this is that we could squeeze a few more seconds out if we could run our code as Windows is starting services on boot instead of waiting for the userdata script to fire. This worked - however, it did break the AWS SSM Agent as a result as it also starts early. In the installation script included in the repo, we work around this by setting the SSM agent startup type to Automatic (Delayed). Faster Configure Network/Start K8s Services This basically involved translating the remainder of the code in the standard AWS bootstrapping scripts in EKS-StartupTask.ps1 Setup Container Network
        ... 
           [DllImport("vmcompute.dll")]
            static extern void HNSCall([MarshalAs(UnmanagedType.LPWStr)] string method,
                           [MarshalAs(UnmanagedType.LPWStr)] string path,
                           [MarshalAs(UnmanagedType.LPWStr)] string request,
                           [MarshalAs(UnmanagedType.LPWStr)] out string response);
            HNSCall("POST", "/networks", jsonString, out response);
        ...

https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L514 Add routes to created NIC
        ...
          routeAddCommands.Append($"route ADD {ipAddrs[i]} MASK 255.255.255.255 0.0.0.0 IF {vNICIndex}");
        ...
        Process process = new Process();
        process.StartInfo.FileName = "cmd.exe";
        process.StartInfo.Arguments = $"/C {routeAddCommands}";
        process.StartInfo.RedirectStandardOutput = true;
        process.StartInfo.UseShellExecute = false;
        process.StartInfo.CreateNoWindow = true;
        process.Start();
        await process.WaitForExitAsync();

https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L586


Start Kube-proxy and Kubelet in parallel once all configuration files have been written (all the previously converted code) Installation To install the bootstrapper, add this to a component in Image Builder: - name: InstallEksWindowsBootstrapper
        action: ExecutePowerShell
        inputs:
          commands:
            - |
              Invoke-WebRequest -Uri 'https://github.com/atg-cloudops/eks-windows-bootstrapper/releases/download/v1.29.1/Install-Service.ps1' -OutFile 'Install-Service.ps1'; 
              .\Install-Service.ps1; 
              Remove-Item 'Install-Service.ps1'; Putting everything together, I tested the output image using these settings: EC2 Instance

m4.xlarge (4 Core)
EBS


GP3 - 120GB/600Mbps/3600 IOPS Testing on these settings got me start times around 55s! We are finally under a minute, maybe spot instances are now viable for Windows EKS Nodes now that the 5-minute penalty is now 1 minute. Hold On… You Said A 30s Start Time! You Said as Fast as Linux! A 30s start time is possible! It just costs a lot. 😢 I ended up testing how fast the sum of all the software was by eliminating as many hardware bottlenecks as I could. Switched from GP3 → IO2 30K IOPS Switched from m4.xlarge → c6a.4xlarge Require Nitro Hypervisor It’s quite easy to change in Karpenter; simply change the Ec2NodeClass to change storage options, and then add the relevant keys in NodePool, e.g. blockDeviceMappings:
    - deviceName: /dev/sda1
      ebs:
        iops: 30000
        volumeSize: 120Gi
        volumeType: io2 - key: karpenter.k8s.aws/instance-hypervisor
          operator: In
          values:
            - 'nitro'
        - key: karpenter.k8s.aws/instance-cpu
          operator: In
          values:
            - '8' The beast that was produced by Karpenter was visible in K8s in 32s. It would be sub-30s with antivirus disabled. Our final tuning produces nodes in around 45s (Using GP3). You can tune for cost/performance yourself for your own scenario. Looking for Help EKS Windows bootstrapper is a simple app that writes config and starts Kubernetes services. It is currently set up for EKS 1.29. As Kubernetes goes through 1.30, 1.31, 1.32… it will need to be kept up to date… I am looking for help with this. 🙂 Previously… https://awstip.com/reducing-eks-windows-node-5-min-start-time-to-90s-eks-windows-bootstrapper-4314f72367a0?embedable=true https://awstip.com/reducing-eks-windows-node-5-min-start-time-to-90s-eks-windows-bootstrapper-4314f72367a0?embedable=true I threw in some tricks I used to make our Windows EKS nodes start in 90s rather than the 5 minutes a standard EKS Windows node takes. I introduced you to the git repo eks-windows-bootstrapper - the application replacing the AWS-provided configuration scripts for EKS Windows. https://github.com/atg-cloudops/eks-windows-bootstrapper?embedable=true https://github.com/atg-cloudops/eks-windows-bootstrapper?embedable=true In today’s story, I will walk you through how to make your Windows nodes take ~30s to start! Recap In the previous story, we got to the point where our start times looked like this: Let’s take a closer look at our latencies now: In order for us to get better launch times, we now have 2 areas where we can improve timing Windows Startup Windows Startup Configure Network/Start K8s Services Configure Network/Start K8s Services Improving Windows Startup Performance As we are using the AWS Image Builder ( https://aws.amazon.com/image-builder/ ) to build our custom AMIs as discussed in the previous story, any changes we make to our image need to be made here. Outside AWS Fast Launch - there really isn’t much we can do to help with the Windows boot process, if you are able - you can uninstall Windows Defender which will massively improve start performance. https://aws.amazon.com/image-builder/ Side note - I found in our use case we do not actually need Fast Launch as we do not actually need to run the OOBE process. We can set our culture using Set-Culture , etc. inside one of our components, and as long as we remove the sections in Unattend.xml related to OOBE, Windows will skip much of the setup process that EC2 Fast Launch deals with. When you start the output AMI, it will start as quickly as a Fast Launch AMI, but without the added expense of pre-provisioning EBS snapshots. Set-Culture Unattend.xml The downside to this approach is that you won’t be able to configure Windows settings on the first boot, what you bake into the AMI is what you will start with - So make sure you set culture/timezone information correctly somewhere in one of your Image Builder components. For reference: So with all avenues to make Windows do less stuff on boot exhausted, there was only one choice left… Changes were made to the EKS Windows Bootstrapper code so it could be run as a Windows Service. The idea behind this is that we could squeeze a few more seconds out if we could run our code as Windows is starting services on boot instead of waiting for the userdata script to fire. This worked - however, it did break the AWS SSM Agent as a result as it also starts early. In the installation script included in the repo, we work around this by setting the SSM agent startup type to Automatic (Delayed) . userdata Automatic (Delayed) Faster Configure Network/Start K8s Services This basically involved translating the remainder of the code in the standard AWS bootstrapping scripts in EKS-StartupTask.ps1 EKS-StartupTask.ps1 Setup Container Network
        ... 
           [DllImport("vmcompute.dll")]
            static extern void HNSCall([MarshalAs(UnmanagedType.LPWStr)] string method,
                           [MarshalAs(UnmanagedType.LPWStr)] string path,
                           [MarshalAs(UnmanagedType.LPWStr)] string request,
                           [MarshalAs(UnmanagedType.LPWStr)] out string response);
            HNSCall("POST", "/networks", jsonString, out response);
        ...

https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L514 Setup Container Network ... 
           [DllImport("vmcompute.dll")]
            static extern void HNSCall([MarshalAs(UnmanagedType.LPWStr)] string method,
                           [MarshalAs(UnmanagedType.LPWStr)] string path,
                           [MarshalAs(UnmanagedType.LPWStr)] string request,
                           [MarshalAs(UnmanagedType.LPWStr)] out string response);
            HNSCall("POST", "/networks", jsonString, out response);
        ... https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L514 Setup Container Network ... 
           [DllImport("vmcompute.dll")]
            static extern void HNSCall([MarshalAs(UnmanagedType.LPWStr)] string method,
                           [MarshalAs(UnmanagedType.LPWStr)] string path,
                           [MarshalAs(UnmanagedType.LPWStr)] string request,
                           [MarshalAs(UnmanagedType.LPWStr)] out string response);
            HNSCall("POST", "/networks", jsonString, out response);
        ... ... 
           [DllImport("vmcompute.dll")]
            static extern void HNSCall([MarshalAs(UnmanagedType.LPWStr)] string method,
                           [MarshalAs(UnmanagedType.LPWStr)] string path,
                           [MarshalAs(UnmanagedType.LPWStr)] string request,
                           [MarshalAs(UnmanagedType.LPWStr)] out string response);
            HNSCall("POST", "/networks", jsonString, out response);
        ... https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L514 https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L514 Add routes to created NIC
        ...
          routeAddCommands.Append($"route ADD {ipAddrs[i]} MASK 255.255.255.255 0.0.0.0 IF {vNICIndex}");
        ...
        Process process = new Process();
        process.StartInfo.FileName = "cmd.exe";
        process.StartInfo.Arguments = $"/C {routeAddCommands}";
        process.StartInfo.RedirectStandardOutput = true;
        process.StartInfo.UseShellExecute = false;
        process.StartInfo.CreateNoWindow = true;
        process.Start();
        await process.WaitForExitAsync();

https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L586 Start Kube-proxy and Kubelet in parallel once all configuration files have been written (all the previously converted code) Add routes to created NIC ...
          routeAddCommands.Append($"route ADD {ipAddrs[i]} MASK 255.255.255.255 0.0.0.0 IF {vNICIndex}");
        ...
        Process process = new Process();
        process.StartInfo.FileName = "cmd.exe";
        process.StartInfo.Arguments = $"/C {routeAddCommands}";
        process.StartInfo.RedirectStandardOutput = true;
        process.StartInfo.UseShellExecute = false;
        process.StartInfo.CreateNoWindow = true;
        process.Start();
        await process.WaitForExitAsync(); https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L586 Add routes to created NIC ...
          routeAddCommands.Append($"route ADD {ipAddrs[i]} MASK 255.255.255.255 0.0.0.0 IF {vNICIndex}");
        ...
        Process process = new Process();
        process.StartInfo.FileName = "cmd.exe";
        process.StartInfo.Arguments = $"/C {routeAddCommands}";
        process.StartInfo.RedirectStandardOutput = true;
        process.StartInfo.UseShellExecute = false;
        process.StartInfo.CreateNoWindow = true;
        process.Start();
        await process.WaitForExitAsync(); ...
          routeAddCommands.Append($"route ADD {ipAddrs[i]} MASK 255.255.255.255 0.0.0.0 IF {vNICIndex}");
        ...
        Process process = new Process();
        process.StartInfo.FileName = "cmd.exe";
        process.StartInfo.Arguments = $"/C {routeAddCommands}";
        process.StartInfo.RedirectStandardOutput = true;
        process.StartInfo.UseShellExecute = false;
        process.StartInfo.CreateNoWindow = true;
        process.Start();
        await process.WaitForExitAsync(); https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L586 https://github.com/atg-cloudops/eks-windows-bootstrapper/blob/8f519953209e3caea590652832acc74585ab1a45/BootstrapperService.cs#L586 Start Kube-proxy and Kubelet in parallel once all configuration files have been written (all the previously converted code) Start Kube-proxy and Kubelet in parallel once all configuration files have been written (all the previously converted code) Installation To install the bootstrapper, add this to a component in Image Builder: - name: InstallEksWindowsBootstrapper
        action: ExecutePowerShell
        inputs:
          commands:
            - |
              Invoke-WebRequest -Uri 'https://github.com/atg-cloudops/eks-windows-bootstrapper/releases/download/v1.29.1/Install-Service.ps1' -OutFile 'Install-Service.ps1'; 
              .\Install-Service.ps1; 
              Remove-Item 'Install-Service.ps1'; - name: InstallEksWindowsBootstrapper
        action: ExecutePowerShell
        inputs:
          commands:
            - |
              Invoke-WebRequest -Uri 'https://github.com/atg-cloudops/eks-windows-bootstrapper/releases/download/v1.29.1/Install-Service.ps1' -OutFile 'Install-Service.ps1'; 
              .\Install-Service.ps1; 
              Remove-Item 'Install-Service.ps1'; Putting everything together, I tested the output image using these settings: EC2 Instance

m4.xlarge (4 Core)
EBS


GP3 - 120GB/600Mbps/3600 IOPS EC2 Instance m4.xlarge (4 Core)
EBS


GP3 - 120GB/600Mbps/3600 IOPS m4.xlarge (4 Core) EBS


GP3 - 120GB/600Mbps/3600 IOPS m4.xlarge (4 Core) EBS GP3 - 120GB/600Mbps/3600 IOPS GP3 - 120GB/600Mbps/3600 IOPS GP3 - 120GB/600Mbps/3600 IOPS GP3 - 120GB/600Mbps/3600 IOPS Testing on these settings got me start times around 55s! We are finally under a minute, maybe spot instances are now viable for Windows EKS Nodes now that the 5-minute penalty is now 1 minute. Hold On… You Said A 30s Start Time! You Said as Fast as Linux! A 30s start time is possible! It just costs a lot. 😢 I ended up testing how fast the sum of all the software was by eliminating as many hardware bottlenecks as I could. Switched from GP3 → IO2 30K IOPS Switched from m4.xlarge → c6a.4xlarge Require Nitro Hypervisor It’s quite easy to change in Karpenter; simply change the Ec2NodeClass to change storage options, and then add the relevant keys in NodePool, e.g. blockDeviceMappings:
    - deviceName: /dev/sda1
      ebs:
        iops: 30000
        volumeSize: 120Gi
        volumeType: io2 blockDeviceMappings:
    - deviceName: /dev/sda1
      ebs:
        iops: 30000
        volumeSize: 120Gi
        volumeType: io2 - key: karpenter.k8s.aws/instance-hypervisor
          operator: In
          values:
            - 'nitro'
        - key: karpenter.k8s.aws/instance-cpu
          operator: In
          values:
            - '8' - key: karpenter.k8s.aws/instance-hypervisor
          operator: In
          values:
            - 'nitro'
        - key: karpenter.k8s.aws/instance-cpu
          operator: In
          values:
            - '8' The beast that was produced by Karpenter was visible in K8s in 32s. It would be sub-30s with antivirus disabled. Our final tuning produces nodes in around 45s (Using GP3). You can tune for cost/performance yourself for your own scenario. Looking for Help EKS Windows bootstrapper is a simple app that writes config and starts Kubernetes services. It is currently set up for EKS 1.29. As Kubernetes goes through 1.30, 1.31, 1.32… it will need to be kept up to date… I am looking for help with this. 🙂

OATFI

oatfinancial.com

-- employees

Since n.d.

COMPANY RANKING

OATFI

EVERGREEN INDEX #7440

Representing this company? Click here to claim and customize this page!

OatFi WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

OATFI

oatfinancial.com

-- employees

Since n.d.

COMPANY RANKING

OATFI

EVERGREEN INDEX #7440

RELATED COMPANIESRANK

STORIES ABOUT OATFI

Representing this company? Click here to claim and customize this page!

OatFi WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!