Cityscope

cityscope

Soviet architectural heritage is very disgusting and romantic at the same time. In every city, from Vladivostok to Kaliningrad, there are the same plattenbau buildings. This makes a lot of people feel this panel architecture a part of their cultural code and nature — depressing and chthonic. Here in Russia, we even have a popular slogan “Russia’s for Sad” (which is also wordplay with right-wings “Russia’s for Russian”). People produce a lot of sad content (music, footage, art) based on these architectural references. So once I thought: what if a neural network would be trained to paint this Sad Russia. What if a soulless algorithm may draw landscapes that are soul pleasure for a generation of Russian doomers. That’s how the was born. project Neurussia Part 1: Data mining The choice of a data source fell on the largest Russian social network — VK. And here is why: VK has a post suggestion mechanism, where people can suggest their own content to be published in large groups. This leads VK to contain a lot more thematic visual content than other social networks. For example, contains around 200 thousand images! Here is what I got initially from VK: the largest group about the aesthetics of Russian ghettos For data scrapping, I used a . In case one would like to reproduce my pipeline, , and below is an example of how to download images from the latest group post. python wrapper for VK API here are API methods descriptions os vk_api urllib.request ( )
    key = input( ) captcha.try_again(key) code = input( ) (code, ) attachment post[ ]: attachment[ ] == :
            photo = attachment[ ] size photo[ ]: size[ ] == imres:
                    url = size[ ]
                    filename = ( .join(url.split( )[ :])).split( )[ ]
                    urllib.request.urlretrieve(url, filename)

phone = input( )
password = input( )
domain = sess = vk_api.VkApi(phone, password, captcha_handler=captcha_handler, auth_handler=auth_handler)
sess.auth()
api = sess.get_api()

posts = api.wall.get(domain=domain, count= )[ ]
save_post_pictures(posts[ ], ) import import import : def captcha_handler (captcha) print f"url: \n" {captcha.get_url()} "Enter captcha code: " return : def auth_handler () "Enter 2FA code: " return True : def save_post_pictures (post, imres) for in 'attachments' if 'type' 'photo' 'photo' for in 'sizes' if 'type' 'url' '_' '/' -2 '?' 0 "phone " "password " 'yebenya' # vk.com/yebenya 1 'items' 0 'z' # https://vk.com/dev/photo_sizes - z max size To filter downloaded images I used and . The idea of using a NN detector is to find all objects that it can find and throw out those images leaving only landscapes footage. Natural Language Toolkit helped me to filter out posts by their captions using . Finally, I got around 10k images of Russian landscapes. EfficientDet NLTK stemming Part 2: StyleGAN training I used the famous Nvidia architecture StyleGAN2 with discriminator augmentations. Authors claim that augmentations allow better results on small datasets (under 30k images). the official code, and my “duct tape” modification for Colab training and video generation (will expand on it below). Here is here is The official paper implementation uses the .tfrecord format to store data for training (how to convert images to .tfrecord is described ). When running on Colab this feature can drastically reduce the number of images that you could process because the official code doesn’t use any compression in tfrecords. here However, that uses compression and probably should allow you to work with larger datasets on Colab (I didn’t try it yet). there is another modification On Colab, each epoch is taking about half an hour for a dataset of 10k images in 1024 resolution. Fun fact: on the very first epochs the colormap has rapidly transformed from random bright colors into 50-shades-of-gray (above picture). After a few epochs, I got something like this: Part 3: Video generation When the first impressive results have achieved an idea to create a music video for my friends-musicians emerged. But how to make the video at least a bit consistent with music? Here the latent space of the generative model may help. The thing is, during training, the neural network has learned how to describe a whole picture with a vector of dimensionality 512. And all the parameters of the final painting are embedded in this vector. So one may find a combination of those parameters that correspond to, for example, time of day, or height of the building, or the number of windows. This may be a complex task itself (however, ), so for my purposes, I decided to simply interpolate between images keyframe-to-keyframe using their latent vectors. For this purpose, I changed only a few lines of code in to make it read a list of keyframes with corresponding latent vectors and generate N intermediate frames linearly interpolating between them with 60fps. there are some approaches generate.py Conclusion With the advent of such approaches as StyleGAN, generative art received a new round of growth. Some are even without writing a line of code. Also, Colab drastically reduces the entry-level and the amount of time to make something interesting work. Hope that this article also will help someone to expand their creativity using modern AI methods. selling generative art on Christie’s auction Resources StyleGAN2-ada VK API wrapper StyleGAN2 modification for Colab Neurussia on instagram https://ferluht.medium.com/neurussia-neural-network-paints-depressing-russian-cityscapes-1052aeb51ae3 Also published on:

This Neural Network Paints Depressing Russian Cityscapes

Uni-OVSeg: A Step Towards Efficient and Bias-Resilient Vision Systems

Authors:
(1) Zhaoqing Wang, The University of Sydney and AI2Robotics;
(2) Xiaobo Xia, The University of Sydney;
(3) Ziye Chen, The University of Melbourne;
(4) Xiao He, AI2Robotics;
(5) Yandong Guo, AI2Robotics;
(6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence;
(7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction 2. Related works 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.2. Main results 4.3. Ablation study 5. Conclusion 6. Broader impacts and References A. Framework details B. Promptable segmentation C. Visualisation 4.2. Main results Open-vocabulary semantic segmentation. As demonstrated in Tab. 1, our study presents a comprehensive comparison of our Uni-OVSeg against previous works across a range of benchmarks. These include ADE20K (encompassing both 150 and 847 class variants), PASCAL Context (459 and 59 class variants), PASCAL VOC (with 20 and 21 class categories), and Cityscapes. Compared to weakly-supervised methods, Uni-OVSeg exhibits remarkable performance improvements across all evaluated datasets. Specifically, in the more challenging datasets of PASCAL Context-459, Uni-OVSeg not only surpasses its weakly-supervised counterparts but also outperforms the cutting-edge fully-supervised methods, e.g., FC-CLIP. This is indicative of Uni-OVSeg’s superior capability in categorizing a diverse array of semantic classes. Furthermore, in the PASCAL VOC benchmarks (20 and 21 classes), Uni-OVSeg demonstrates a substantial enhancement over state-of-the-art weakly-supervised methods, achieving improvements of 18.3% and 12.2% mIoU, respectively, which demonstrates our Uni-OVSeg captures fine-grained spatial structures. These results elevate the practical applicability of weakly-supervised open-vocabulary segmentation to new heights. Open-vocabulary panoptic segmentation. For openvocabulary panoptic segmentation, we zero-shot evaluate our model on the COCO, ADE20K, and Cityscapes datasets, shown in Tab. 2. Existing weakly-supervised methods only use text supervision, which causes a challenge to discriminate different instances with the same semantic class. To the best of our knowledge, we are the first to learn open-vocabulary panoptic segmentation with weak supervision. Compared to unsupervised methods, we obviously outperform U2Seg by 1.9% PQ, 1.5%SQ, and 4.4% RQ on COCO datasets. Unfortunately, our used image-mask pairs contain multiple granularity masks, such as object-wise and part-wise masks, which is different with the panoptic segmentation datasets. This discrepancy leads to a number of false positive results, limiting our performance. Promptable segmentation. To evaluate the segmentation efficacy of Uni-OVSeg in scenarios involving interactive point and box prompts, we conduct comparative analyses with the SAM-ViT/L model across a range of datasets from diverse domains. For visual prompts, we implement a uniform 20×20 point grid as the interactive point prompt and utilise the actual bounding boxes as box prompts. The segmentation performance is measured using the 1-pt IoU (Oracle) metric across all datasets. The results, as demonstrated in Fig. 3, indicate that Uni-OVSeg exceeds SAM in most datasets when employing point and box prompts. In particular, Uni-OVSeg significantly surpasses SAM in terms of IoU by 31.1%, 20.1%, and 43.2% on the TimberSeg, COCO instance, and PIDray datasets, respectively. Additionally, we provide a visualisation of the segmentation results of both SAM and Uni-OVSeg in Fig. 4. The visualisation reveals that SAM tends to miss larger masks and erroneously categorise areas with unclear boundaries. In contrast, UniOVSeg demonstrates exceptional precision in segmenting objects of varying scales and different stuff. More experimental results are visualisation can be found in the Appendix B and C. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Authors: Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related works 2. Related works 3. Method and 3.1. Problem definition 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.1. Implementation details 4.2. Main results 4.2. Main results 4.3. Ablation study 4.3. Ablation study 5. Conclusion 5. Conclusion 6. Broader impacts and References 6. Broader impacts and References A. Framework details A. Framework details B. Promptable segmentation B. Promptable segmentation C. Visualisation C. Visualisation 4.2. Main results Open-vocabulary semantic segmentation. As demonstrated in Tab. 1, our study presents a comprehensive comparison of our Uni-OVSeg against previous works across a range of benchmarks. These include ADE20K (encompassing both 150 and 847 class variants), PASCAL Context (459 and 59 class variants), PASCAL VOC (with 20 and 21 class categories), and Cityscapes. Compared to weakly-supervised methods, Uni-OVSeg exhibits remarkable performance improvements across all evaluated datasets. Specifically, in the more challenging datasets of PASCAL Context-459, Uni-OVSeg not only surpasses its weakly-supervised counterparts but also outperforms the cutting-edge fully-supervised methods, e.g., FC-CLIP. This is indicative of Uni-OVSeg’s superior capability in categorizing a diverse array of semantic classes. Furthermore, in the PASCAL VOC benchmarks (20 and 21 classes), Uni-OVSeg demonstrates a substantial enhancement over state-of-the-art weakly-supervised methods, achieving improvements of 18.3% and 12.2% mIoU, respectively, which demonstrates our Uni-OVSeg captures fine-grained spatial structures. These results elevate the practical applicability of weakly-supervised open-vocabulary segmentation to new heights. Open-vocabulary semantic segmentation. Open-vocabulary panoptic segmentation. For openvocabulary panoptic segmentation, we zero-shot evaluate our model on the COCO, ADE20K, and Cityscapes datasets, shown in Tab. 2. Existing weakly-supervised methods only use text supervision, which causes a challenge to discriminate different instances with the same semantic class. To the best of our knowledge, we are the first to learn open-vocabulary panoptic segmentation with weak supervision. Compared to unsupervised methods, we obviously outperform U2Seg by 1.9% PQ, 1.5%SQ, and 4.4% RQ on COCO datasets. Unfortunately, our used image-mask pairs contain multiple granularity masks, such as object-wise and part-wise masks, which is different with the panoptic segmentation datasets. This discrepancy leads to a number of false positive results, limiting our performance. Open-vocabulary panoptic segmentation. Promptable segmentation. To evaluate the segmentation efficacy of Uni-OVSeg in scenarios involving interactive point and box prompts, we conduct comparative analyses with the SAM-ViT/L model across a range of datasets from diverse domains. For visual prompts, we implement a uniform 20×20 point grid as the interactive point prompt and utilise the actual bounding boxes as box prompts. The segmentation performance is measured using the 1-pt IoU (Oracle) metric across all datasets. The results, as demonstrated in Fig. 3, indicate that Uni-OVSeg exceeds SAM in most Promptable segmentation. datasets when employing point and box prompts. In particular, Uni-OVSeg significantly surpasses SAM in terms of IoU by 31.1%, 20.1%, and 43.2% on the TimberSeg, COCO instance, and PIDray datasets, respectively. Additionally, we provide a visualisation of the segmentation results of both SAM and Uni-OVSeg in Fig. 4. The visualisation reveals that SAM tends to miss larger masks and erroneously categorise areas with unclear boundaries. In contrast, UniOVSeg demonstrates exceptional precision in segmenting objects of varying scales and different stuff. More experimental results are visualisation can be found in the Appendix B and C. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Uni-OVSeg Outperforms Weakly-Supervised and Fully-Supervised Methods in Open-Vocabulary Segmentation

Authors:
(1) Zhaoqing Wang, The University of Sydney and AI2Robotics;
(2) Xiaobo Xia, The University of Sydney;
(3) Ziye Chen, The University of Melbourne;
(4) Xiao He, AI2Robotics;
(5) Yandong Guo, AI2Robotics;
(6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence;
(7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction 2. Related works 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.2. Main results 4.3. Ablation study 5. Conclusion 6. Broader impacts and References A. Framework details B. Promptable segmentation C. Visualisation 4.1. Implementation details Datasets. During training, we randomly sample the 30% subset from the SA-1B dataset [34], which contains ∼ 3 million images and ∼ 0.3 billion masks. Although this supervision provides diverse binary masks, it lacks the semantic class for each mask. In addition, following Chen et al. [8], we collect about 1.3 million image-text pairs and use a large vision-language model to refine them. Afterward, we use the ChatGPT-based parser to extract entities with descriptive words from these text descriptions. Evaluation & metrics. We evaluate our model mainly on three tasks, including open-vocabulary semantic segmentation, open-vocabulary panoptic segmentation, and promptable segmentation. Following previous work [76], we adopt prompt engineering from [21, 66] and prompt templates from [22, 37]. For open-vocabulary semantic segmentation, we zero-shot evaluate the model on the COCO [40], ADE20K [84], PASCAL [18] datasets. The open-vocabulary semantic segmentation results are evaluated with the mean Intersection-over-Union (mIoU). For open-vocabulary panoptic segmentation, we evaluate the model on the COCO, ADE20K, and Cityscapes [15] datasets. We report the panoptic quality (PQ), semantic quality (SQ), and recognition quality (RQ) for open-vocabulary panoptic segmentation. For promptable segmentation, we report the 1-Point and 1-Box IoU (Oracle) on a wide range of datasets. Oracle denotes that we select the output mask with the max IoU by calculating the IoU between the prediction and target mask. More details can be found in Appendix B. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Authors: Authors: (1) Zhaoqing Wang, The University of Sydney and AI2Robotics; (2) Xiaobo Xia, The University of Sydney; (3) Ziye Chen, The University of Melbourne; (4) Xiao He, AI2Robotics; (5) Yandong Guo, AI2Robotics; (6) Mingming Gong, The University of Melbourne and Mohamed bin Zayed University of Artificial Intelligence; (7) Tongliang Liu, The University of Sydney. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Related works 2. Related works 3. Method and 3.1. Problem definition 3. Method and 3.1. Problem definition 3.2. Baseline and 3.3. Uni-OVSeg framework 3.2. Baseline and 3.3. Uni-OVSeg framework 4. Experiments 4.1. Implementation details 4.1. Implementation details 4.2. Main results 4.2. Main results 4.3. Ablation study 4.3. Ablation study 5. Conclusion 5. Conclusion 6. Broader impacts and References 6. Broader impacts and References A. Framework details A. Framework details B. Promptable segmentation B. Promptable segmentation C. Visualisation C. Visualisation 4.1. Implementation details Datasets . During training, we randomly sample the 30% subset from the SA-1B dataset [34], which contains ∼ 3 million images and ∼ 0.3 billion masks. Although this supervision provides diverse binary masks, it lacks the semantic class for each mask. In addition, following Chen et al. [8], we collect about 1.3 million image-text pairs and use a large vision-language model to refine them. Afterward, we use the ChatGPT-based parser to extract entities with descriptive words from these text descriptions. Datasets Evaluation & metrics. We evaluate our model mainly on three tasks, including open-vocabulary semantic segmentation, open-vocabulary panoptic segmentation, and promptable segmentation. Following previous work [76], we adopt prompt engineering from [21, 66] and prompt templates from [22, 37]. For open-vocabulary semantic segmentation, we zero-shot evaluate the model on the COCO [40], ADE20K [84], PASCAL [18] datasets. The open-vocabulary semantic segmentation results are evaluated with the mean Intersection-over-Union (mIoU). For open-vocabulary panoptic segmentation, we evaluate the model on the COCO, ADE20K, and Cityscapes [15] datasets. We report the panoptic quality (PQ), semantic quality (SQ), and recognition quality (RQ) for open-vocabulary panoptic segmentation. For promptable segmentation, we report the 1-Point and 1-Box IoU (Oracle) on a wide range of datasets. Oracle denotes that we select the output mask with the max IoU by calculating the IoU between the prediction and target mask. More details can be found in Appendix B. Evaluation & metrics. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Datasets and Evaluation Methods for Open-Vocabulary Segmentation Tasks

Have you ever found yourself staring at an image spotted somewhere on the internet and wondering whether it results from human or artificial creative labor? Since the breakthrough in image generation technology, I’ve been captivated by interior designs, nature scenery, fashion photographs, and futuristic cityscapes made with AI that look like professional artwork. As someone with a creative eye yet no artistic skills, I was excited about the opportunity to transform ideas from my mind into visuals based solely on text input. I tried out several popular text-to-image tools, like MidJourney and DreamStudio on the web and ARTA on mobile. Still, while they all provided a good starting point and a straightforward user experience, the generation outcomes often fell short of my expectations and differed from what I envisioned — until I learned to tell AI what to draw in the language it understands. If you’re new to AI-generated art and need assistance streamlining the creative process, I’ll provide a step-by-step guide below to help you compose concise and well-structured text prompts that the AI models can effectively interpret and translate into the desired images. In the bottom line, I will share an example of both a correct and an incorrect prompt to illustrate how significantly they can impact the generation output in practice. First things first, two common prompt-writing mistakes you should avoid when using any text-to-image generator are being too wordy and using complex grammar. In fact, the more words you incorporate and the more complicated your grammar, the more difficult it is for AI to understand your prompt and adequately depict your concept. Therefore, communicate concisely, carefully choosing keywords, and list your ideas explicitly and separately rather than trying to cram them into one sentence. Step 1. Choose the Content Type Ask yourself how you want your image to look overall. Do you see it as a photograph, illustration, painting, sketch, or other artwork type? This initial decision puts you in control of the final result and helps align all the elements with this general framework. For instance, choosing a sketch will typically produce pencil images, while opting for a painting will add colors and likely create a more vintage appearance. Step 2. Define the Subject AI image generators require a specific subject for rendering, so you must indicate the main subject of your future image, whether a person, an animal, or a landscape. Stick to one subject per prompt and opt for concrete nouns like girl, dog, train, rose, ocean, temple, etc., for more accurate results. You can most likely render something using abstract keywords like love, peace, or joy, but the results may appear very inconsistent in what they depict. Step 3. Add Subject Description Now, it’s time to detail the concept with modifiers – adjectives or attributive nouns that add to the sense of a head subject. Modifiers can completely change the perspective of your image and boost its overall quality, so the more descriptive you are, the better the generated output. Define what the subject looks like, what the subject is doing, and other characteristics essential to representing your concept. Step 4. Describe the Context To help AI create the exact atmosphere and surroundings you need, give attention to aspects like lightning (studio, soft, cinematic, volumetric, accent lighting, sunlight, frontlight, backlight, ambient, moonlight, etc.), color theme (pastel, vibrant, neon pink, dark, etc.), environment/scene (in the sky, indoor, underwater, at night, etc.), and background (forest, beach, solid color, nebula, etc.). Additionally, consider the mood and vibe you want to convey. Simple feeling modifiers can set the scene’s atmosphere, whether you opt for positive (energetic, cozy, romantic, satisfaction, etc.) or negative (depressive, loneliness, regret, fear, etc.) ones. Step 5. Define Perspective Add a point of view to determine the angle from which you are looking at the image, such as side, overhead, or front. Other essential modifiers for creating photo-like images involve lenses and cameras. Here are a few examples: close-up, macro, lens flare, wide angle, microscopic, bokeh, isometric, depth of field, ultra-wide angle, fisheye lens, and panorama. Step 6. Add Art Style, Art Mediums, and Artists Proceed to aesthetic settings and define the overall artistic style. Also, specify the art medium. These aspects can greatly contribute to the value of the generated output, making it more consistent and expressive. Here are some of the common keywords: geometrical art, pop art, watercolor, line drawing, line art, 3D sculpture, crayon drawing, painting, oil painting, pastel drawing, medieval art, photography, digital art, stained glass, cave painting, polaroid, vector art, retro, illustration, comics, editorial fashion photography, Japanese anime style, surreal, travel photography, pencil sketch, steampunk, cartoon, futurism, vaporwave, art deco, interior design, psychedelic, street art. Consider including artistic influences. Adding the name of a specific artist whose signature artwork aligns with your chosen art genre can enhance the final image. Note, however, that finding the right artist is crucial so as not to mess up the output. Some artists affect the generation only subtly, while others can alter it significantly depending on their distinctive manner. For example, Claude Monet can infuse impressionism into landscapes, while mentioning Salvador Dali adds a surrealistic touch. Step 7. Add “Magic” Words Certain words can make a dramatic difference in the outcomes of the generation. The keywords like highly detailed, ultra-realistic, 4K, 8K, UHD, and HDR are the game changers for the overall quality of the image. Adding “studio lighting” will introduce appealing textures to the image, while “accent lighting” can draw attention to specific features. Describing the image as “professional” can enhance color contrast and detail, while mentioning “vivid colors” will bring life to your image. Using terms like ultra-wide angle, panoramic, or long-shot will help guide the camera. Packing your prompt with phrases related to artistic techniques and materials (such as trending on artstation, precise line-art, golden ratio, fine art, etc.) and those associated with artistic value (such as most beautiful image ever seen, epic composition, etc.) can also bring sophisticated improvements. Step 8. Consider Adding Negative Prompts Negative prompts are parameters that inform the generation model about what you do not want to see in the final image. They help prevent AI from generating specific elements, fix image abnormalities, and improve overall quality. Here are some universal examples: blurry, cropped, cloned face, duplicate, error, ugly, wrong proportions, deform, fused fingers, extra arms, law quality res, watermark, and more. Step 9. Define Correct Word Order and Keyword Weight Input The order is almost as important as the vocabulary, guiding the AI in generating the desired output. Combine keywords to direct the focus of your prompt and note key phrases that influence image preference. Emphasize essential keywords with a full colon and decimal number, use parentheses for added importance, and brackets for less importance. Typically, the earlier a word appears in the sentence, the more importance it is given. If the AI model didn’t take in some key phrase and failed to depict it in the generated image, try rearranging the order of words, placing the missing element earlier in the prompt. Bad vs. Good Prompt Example Bad prompt: portrait of a woman looking directly at the camera, neck tattoo, urban pink background. Good prompt: head and shoulders realistic photo portrait of a beautiful woman looking at the camera, white skin, pink lips, red hair, blue eyes, black tattoos on her neck, neon punk, highly detailed, pink neon city background. Creating effective prompts requires both artistic imagination and technical skills. By including the necessary building blocks and following the proper prompt structure you ensure that the text-to-image generation model understands your vision and produces optimal output. However, don’t be afraid to experiment; the more you practice, the more adept you become, and the more remarkable your creations will be. Have you ever found yourself staring at an image spotted somewhere on the internet and wondering whether it results from human or artificial creative labor? Since the breakthrough in image generation technology, I’ve been captivated by interior designs, nature scenery, fashion photographs, and futuristic cityscapes made with AI that look like professional artwork. Have you ever found yourself staring at an image spotted somewhere on the internet and wondering whether it results from human or artificial creative labor? Since the breakthrough in image generation technology, I’ve been captivated by interior designs, nature scenery, fashion photographs, and futuristic cityscapes made with AI that look like professional artwork. As someone with a creative eye yet no artistic skills, I was excited about the opportunity to transform ideas from my mind into visuals based solely on text input. I tried out several popular text-to-image tools, like MidJourney and DreamStudio on the web and ARTA on mobile. Still, while they all provided a good starting point and a straightforward user experience, the generation outcomes often fell short of my expectations and differed from what I envisioned — until I learned to tell AI what to draw in the language it understands. If you’re new to AI-generated art and need assistance streamlining the creative process, I’ll provide a step-by-step guide below to help you compose concise and well-structured text prompts that the AI models can effectively interpret and translate into the desired images. In the bottom line, I will share an example of both a correct and an incorrect prompt to illustrate how significantly they can impact the generation output in practice. First things first, two common prompt-writing mistakes you should avoid when using any text-to-image generator are being too wordy and using complex grammar. In fact, the more words you incorporate and the more complicated your grammar, the more difficult it is for AI to understand your prompt and adequately depict your concept. Therefore, communicate concisely, carefully choosing keywords, and list your ideas explicitly and separately rather than trying to cram them into one sentence. Step 1. Choose the Content Type Ask yourself how you want your image to look overall. Do you see it as a photograph, illustration, painting, sketch, or other artwork type? This initial decision puts you in control of the final result and helps align all the elements with this general framework. For instance, choosing a sketch will typically produce pencil images, while opting for a painting will add colors and likely create a more vintage appearance. Step 2. Define the Subject AI image generators require a specific subject for rendering, so you must indicate the main subject of your future image, whether a person, an animal, or a landscape. Stick to one subject per prompt and opt for concrete nouns like girl, dog, train, rose, ocean, temple, etc., for more accurate results. You can most likely render something using abstract keywords like love, peace, or joy, but the results may appear very inconsistent in what they depict. Step 3. Add Subject Description Now, it’s time to detail the concept with modifiers – adjectives or attributive nouns that add to the sense of a head subject. Modifiers can completely change the perspective of your image and boost its overall quality, so the more descriptive you are, the better the generated output. Define what the subject looks like, what the subject is doing, and other characteristics essential to representing your concept. Step 4. Describe the Context To help AI create the exact atmosphere and surroundings you need, give attention to aspects like lightning (studio, soft, cinematic, volumetric, accent lighting, sunlight, frontlight, backlight, ambient, moonlight, etc.), color theme (pastel, vibrant, neon pink, dark, etc.), environment/scene (in the sky, indoor, underwater, at night, etc.), and background (forest, beach, solid color, nebula, etc.). Additionally, consider the mood and vibe you want to convey. Simple feeling modifiers can set the scene’s atmosphere, whether you opt for positive (energetic, cozy, romantic, satisfaction, etc.) or negative (depressive, loneliness, regret, fear, etc.) ones. Step 5. Define Perspective Add a point of view to determine the angle from which you are looking at the image, such as side, overhead, or front. Other essential modifiers for creating photo-like images involve lenses and cameras. Here are a few examples: close-up, macro, lens flare, wide angle, microscopic, bokeh, isometric, depth of field, ultra-wide angle, fisheye lens, and panorama. Step 6. Add Art Style, Art Mediums, and Artists Proceed to aesthetic settings and define the overall artistic style. Also, specify the art medium. These aspects can greatly contribute to the value of the generated output, making it more consistent and expressive. Here are some of the common keywords: geometrical art, pop art, watercolor, line drawing, line art, 3D sculpture, crayon drawing, painting, oil painting, pastel drawing, medieval art, photography, digital art, stained glass, cave painting, polaroid, vector art, retro, illustration, comics, editorial fashion photography, Japanese anime style, surreal, travel photography, pencil sketch, steampunk, cartoon, futurism, vaporwave, art deco, interior design, psychedelic, street art. Consider including artistic influences. Adding the name of a specific artist whose signature artwork aligns with your chosen art genre can enhance the final image. Note, however, that finding the right artist is crucial so as not to mess up the output. Some artists affect the generation only subtly, while others can alter it significantly depending on their distinctive manner. For example, Claude Monet can infuse impressionism into landscapes, while mentioning Salvador Dali adds a surrealistic touch. Step 7. Add “Magic” Words Certain words can make a dramatic difference in the outcomes of the generation. The keywords like highly detailed, ultra-realistic, 4K, 8K, UHD, and HDR are the game changers for the overall quality of the image. Adding “studio lighting” will introduce appealing textures to the image, while “accent lighting” can draw attention to specific features. Describing the image as “professional” can enhance color contrast and detail, while mentioning “vivid colors” will bring life to your image. Using terms like ultra-wide angle, panoramic, or long-shot will help guide the camera. Packing your prompt with phrases related to artistic techniques and materials (such as trending on artstation, precise line-art, golden ratio, fine art, etc.) and those associated with artistic value (such as most beautiful image ever seen, epic composition, etc.) can also bring sophisticated improvements. Step 8. Consider Adding Negative Prompts Negative prompts are parameters that inform the generation model about what you do not want to see in the final image. They help prevent AI from generating specific elements, fix image abnormalities, and improve overall quality. Here are some universal examples: blurry, cropped, cloned face, duplicate, error, ugly, wrong proportions, deform, fused fingers, extra arms, law quality res, watermark, and more. Step 9. Define Correct Word Order and Keyword Weight Input The order is almost as important as the vocabulary, guiding the AI in generating the desired output. Combine keywords to direct the focus of your prompt and note key phrases that influence image preference. Emphasize essential keywords with a full colon and decimal number, use parentheses for added importance, and brackets for less importance. Typically, the earlier a word appears in the sentence, the more importance it is given. If the AI model didn’t take in some key phrase and failed to depict it in the generated image, try rearranging the order of words, placing the missing element earlier in the prompt. Bad vs. Good Prompt Example Bad prompt: portrait of a woman looking directly at the camera, neck tattoo, urban pink background. Good prompt: head and shoulders realistic photo portrait of a beautiful woman looking at the camera, white skin, pink lips, red hair, blue eyes, black tattoos on her neck, neon punk, highly detailed, pink neon city background. Creating effective prompts requires both artistic imagination and technical skills. By including the necessary building blocks and following the proper prompt structure you ensure that the text-to-image generation model understands your vision and produces optimal output. However, don’t be afraid to experiment; the more you practice, the more adept you become, and the more remarkable your creations will be.

How to Write Text Prompts for AI Image Generators to Get Optimal Results with Ease

Beijing, China – RockTree Capital, a leading crypto-native fund and merchant bank headquartered in Beijing, announces the launch of its immersive website, showcasing a futuristic film-noir concept: the Cyberpunk Crypto City. The website’s cutting-edge interactive cityscape represents RockTree's vision where Web3 has become pervasive, fusing technology, finance, and culture. The site, set in the future near the 11th halving of bitcoin, builds on the ideas of the cypherpunk movement, through advocating for the widespread use of decentralization and privacy-enhancing technologies as a route to self sovereignty and freedom for human beings everywhere.  As an early-stage investor and bespoke accelerator in Asia for its portfolio companies, RockTree Capital has been instrumental in geometrically accelerating its investee projects’ go-to-market cycle. https://www.youtube.com/watch?v=NUFuYDJBO4A&embedable=true Through its ultra-localized strategies and grassroots community building initiatives, RockTree has empowered emerging founders and projects to thrive in both East and West markets. Leveraging deep regional relationships with leading crypto institutions, capital collaborators, and Web3 partners, RockTree invests in and accelerates the trajectory of projects from Infrastructure, DeFi, Cross-chain, and the Bitcoin Ecosystem. Founded by visionary investor Omer Ozden and headquartered in Beijing, China, RockTree Capital operates at the nexus of the East & West with a crypto native team that bridges the human, cultural, capital and media gap across these markets. RockTree embodies a truly grassroots approach that unites Eastern and Western crypto markets and benefits its portfolio investments, which include amongst others; dYdX, The Graph, Chainlink, Fantom, Pocket, Maple, Covalent, Axelar, Eclipse, and BEVM. "RockTree focuses on the human aspects of Web3. In addition to unifying the collaboration of top people in Eastern and Western crypto markets, our core strength is identifying the human hallmarks of success for Founders and Team," says Omer Ozden, Founder and Chairman of RockTree Capital. "We utilize advanced behavioral neuroscience methods to determine Founders’ emotional intelligence, execution capabilities and integrity. We prioritize not only technological innovation, but also the individuals behind that innovation, their psychology and levels of consciousness. And we invite those projects into our trusted international ecosystem and methodology, so they can scale rapidly in foreign markets, instead of trying to figure it out organically through trial and error," said Ozden. The Cyberpunk Crypto City website represents RockTree's forward-looking vision decades in the future, where Web3 technology and digital assets penetrate all peoples’ daily lives, and decentralized finance becomes a pervasive consumer product, like soft drinks or chewing gum, allowing true democratization of finance. Visitors to the Cyberpunk Crypto City website will immerse themselves in a retro film-noir homage, with captivating visuals of a cybernetic metropolis of replicants, that is bilingual in English and Chinese, and powered by blockchain technology. Experience the future of finance at RockTree Capital's Cyberpunk Crypto City and join us in shaping the next chapter of Web3 innovation. For media inquiries, please contact: Alisa Loi lisa@rocktreecapital.com About RockTree Capital RockTree Capital is an early stage crypto-native fund and merchant bank based in Beijing, China. Our portfolio companies experience geometric scale in Asian markets through the RockTree Acceleration Program by combining our ultra-localized go-to-market strategies and grassroots long-term community building. RockTree invests into top-tier crypto projects and helps build Decacorns in the areas of Infrastructure, DeFi, Cross-chain and Bitcoin Ecosystem www.rocktreecapital.com https://twitter.com/RockTreeCapital This story was distributed as a release by Btcwire under HackerNoon’s Business Blogging Program. Learn more about the program here. Beijing, China – RockTree Capital, a leading crypto-native fund and merchant bank headquartered in Beijing, announces the launch of its immersive website, showcasing a futuristic film-noir concept: the Cyberpunk Crypto City. The website’s cutting-edge interactive cityscape represents RockTree's vision where Web3 has become pervasive, fusing technology, finance, and culture. The site, set in the future near the 11th halving of bitcoin, builds on the ideas of the cypherpunk movement, through advocating for the widespread use of decentralization and privacy-enhancing technologies as a route to self sovereignty and freedom for human beings everywhere.  As an early-stage investor and bespoke accelerator in Asia for its portfolio companies, RockTree Capital has been instrumental in geometrically accelerating its investee projects’ go-to-market cycle. RockTree Capital RockTree Capital https://www.youtube.com/watch?v=NUFuYDJBO4A&embedable=true https://www.youtube.com/watch?v=NUFuYDJBO4A&embedable=true Through its ultra-localized strategies and grassroots community building initiatives, RockTree has empowered emerging founders and projects to thrive in both East and West markets. Leveraging deep regional relationships with leading crypto institutions, capital collaborators, and Web3 partners, RockTree invests in and accelerates the trajectory of projects from Infrastructure, DeFi, Cross-chain, and the Bitcoin Ecosystem. Founded by visionary investor Omer Ozden and headquartered in Beijing, China, RockTree Capital operates at the nexus of the East & West with a crypto native team that bridges the human, cultural, capital and media gap across these markets. RockTree embodies a truly grassroots approach that unites Eastern and Western crypto markets and benefits its portfolio investments, which include amongst others; dYdX, The Graph, Chainlink, Fantom, Pocket, Maple, Covalent, Axelar, Eclipse, and BEVM. "RockTree focuses on the human aspects of Web3. In addition to unifying the collaboration of top people in Eastern and Western crypto markets, our core strength is identifying the human hallmarks of success for Founders and Team," says Omer Ozden, Founder and Chairman of RockTree Capital. "RockTree focuses on the human aspects of Web3. In addition to unifying the collaboration of top people in Eastern and Western crypto markets, our core strength is identifying the human hallmarks of success for Founders and Team," says Omer Ozden, Founder and Chairman of RockTree Capital. "We utilize advanced behavioral neuroscience methods to determine Founders’ emotional intelligence, execution capabilities and integrity. We prioritize not only technological innovation, but also the individuals behind that innovation, their psychology and levels of consciousness. And we invite those projects into our trusted international ecosystem and methodology, so they can scale rapidly in foreign markets, instead of trying to figure it out organically through trial and error," said Ozden. "We utilize advanced behavioral neuroscience methods to determine Founders’ emotional intelligence, execution capabilities and integrity. We prioritize not only technological innovation, but also the individuals behind that innovation, their psychology and levels of consciousness. And we invite those projects into our trusted international ecosystem and methodology, so they can scale rapidly in foreign markets, instead of trying to figure it out organically through trial and error," said Ozden. The Cyberpunk Crypto City website represents RockTree's forward-looking vision decades in the future, where Web3 technology and digital assets penetrate all peoples’ daily lives, and decentralized finance becomes a pervasive consumer product, like soft drinks or chewing gum, allowing true democratization of finance. Visitors to the Cyberpunk Crypto City website will immerse themselves in a retro film-noir homage, with captivating visuals of a cybernetic metropolis of replicants, that is bilingual in English and Chinese, and powered by blockchain technology. Experience the future of finance at RockTree Capital's Cyberpunk Crypto City and join us in shaping the next chapter of Web3 innovation. RockTree Capital's Cyberpunk Crypto City RockTree Capital's Cyberpunk Crypto City For media inquiries, please contact: Alisa Loi lisa@rocktreecapital.com lisa@rocktreecapital.com lisa@rocktreecapital.com About RockTree Capital RockTree Capital is an early stage crypto-native fund and merchant bank based in Beijing, China. Our portfolio companies experience geometric scale in Asian markets through the RockTree Acceleration Program by combining our ultra-localized go-to-market strategies and grassroots long-term community building. RockTree invests into top-tier crypto projects and helps build Decacorns in the areas of Infrastructure, DeFi, Cross-chain and Bitcoin Ecosystem www.rocktreecapital.com www.rocktreecapital.com www.rocktreecapital.com https://twitter.com/RockTreeCapital https://twitter.com/RockTreeCapital https://twitter.com/RockTreeCapital This story was distributed as a release by Btcwire under HackerNoon’s Business Blogging Program. Learn more about the program here. This story was distributed as a release by Btcwire under HackerNoon’s Business Blogging Program. Learn more about the program here . here here

RockTree Capital Unveils Cyberpunk Crypto Future in a New Website

Advanced Algorithms: Stacks of Cubes Surface Area

82 Stories To Learn About Neural Networks

(AI) is at the forefront, reshaping our digital world and one fascinating aspect is AI-generated 3D models. In this article, we’ll explore this cool tech in simple terms, from what it is to the magic of the "render farm. Artificial intelligence What is an AI-generated 3D Model? Imagine a digital sculptor with an uncanny ability to create intricate 3D designs effortlessly. This digital artist is none other than artificial intelligence. AI-generated 3D models are three-dimensional representations of objects, environments, or characters crafted by intelligent algorithms rather than human hands. These models range from architectural designs to characters in video games, and they're not just static images—they're dynamic, allowing for realistic movements and interactions. The Magic of the Render Farm The Birth of the Model: Rendering Explained Before we delve into the wonders of AI-generated 3D models, let's introduce a crucial player in this digital orchestra—the render farms like the one from . Think of it as a powerhouse of computation, where the magic of rendering takes place. Rendering is the process of turning a 3D model into a 2D image or video. This transformation involves complex calculations of light, shadows, and textures to create the final visual masterpiece. GarageFarm Render Farm: Fueling the Creative Engine The render farm is a network of high-performance computers working together to process the immense amount of data required for rendering. It's like having a team of tireless workers assembling each pixel of an image or frame of a video. This collaborative effort not only accelerates the rendering process but also enables the creation of high-quality visuals that would be impractical for a single machine to handle. AI in Action: Transforming 3D Modeling How AI Enhances 3D Modeling Traditional involves skilled artists meticulously crafting each detail. However, AI injects a dose of automation into this process. Using advanced algorithms, AI can analyze vast datasets and learn to generate 3D models autonomously. This not only speeds up the production pipeline but also opens the doors to new levels of creativity. 3D modeling Infinite Creativity: Endless Possibilities AI-generated 3D models bring forth a new era of creativity. Whether it's designing futuristic cityscapes, otherworldly creatures, or realistic product prototypes, the possibilities are limitless. The ability of AI to understand patterns and aesthetics allows it to create designs that may surprise even the most seasoned human artists. Applications Across Industries From Entertainment to Architecture: Where AI Shines The impact of AI-generated 3D models extends across various industries. In the realm of entertainment, video game developers leverage AI to populate their virtual worlds with diverse and realistic characters. Architects use AI-generated models to visualize and refine their designs, offering clients an immersive preview of their projects. The medical field benefits, too, as AI aids in creating detailed 3D representations for surgical planning and training. The Marriage of AI and Human Creativity A Collaborative Symphony While AI takes the lead in generating 3D models, human creativity remains an indispensable part of the process. Artists and designers work alongside AI algorithms, guiding the system and infusing their unique touch into the creations. This collaborative approach results in a harmonious blend of artificial intelligence and human ingenuity. Challenges and Future Horizons Navigating Challenges As with any technological leap, there are challenges to overcome. Ensuring ethical AI use, addressing biases in algorithms, and refining the balance between automation and human touch are ongoing considerations. However, the dynamic nature of AI means these challenges are met with continuous innovation. The Future: AI-Generated 3D Models Evolving Looking ahead, the evolution of AI-generated 3D models holds great promise. As algorithms become more sophisticated, the realism and diversity of generated models will continue to advance. This not only transforms industries but also democratizes creativity, offering tools for expression to a broader audience. Final Thoughts AI-generated 3D models stand as a testament to the fusion of innovation and creativity. The render farm, a silent hero in this narrative, powers the transformation of algorithms into visual marvels. From entertainment to healthcare, AI's impact is profound, reshaping how we visualize the virtual realm. As we navigate the challenges and embrace the collaborative potential of AI and human creativity, the future promises a digital renaissance where the boundaries of imagination are pushed further than ever before. GIF source:  Art Fire by PurzBeats

Unlocking the Future: AI-Generated 3D Models

ভবিষ্যৎ আনলক করা: এআই-জেনারেটেড 3D মডেল

Die Zukunft erschließen: KI-generierte 3D-Modelle

Liberando el futuro: modelos 3D generados por IA

Ouvrir l'avenir : modèles 3D générés par l'IA

भविष्य को खोलना: एआई-जनरेटेड 3डी मॉडल

未来の扉を開く: AI 生成の 3D モデル

미래를 열다: AI가 생성한 3D 모델

Desbloqueando o futuro: modelos 3D gerados por IA

Открывая будущее: 3D-модели, созданные искусственным интеллектом

Geleceğin Kilidini Açmak: Yapay Zeka Tarafından Oluşturulan 3D Modeller

Mở khóa tương lai: Mô hình 3D do AI tạo ra

解锁未来：AI 生成的 3D 模型

Note: This was originally written in 2019. During his travels, author and podcaster mentioned on comedian that the first thing he likes to do in a new city is to take a bike tour. I followed that move, taking a walking tour my first few days after arriving to get a sense of the areas, following that up with experiencing local marketplaces, bars, and nightlife. Tim Ferriss Ari Shaffir's podcast I traveled to Medellín, Colombia, both to explore for a week on vacation and then to immerse myself in daily life, while working remotely as the creator of , the first cryptocurrency news audiocast app. Coinverse Tours Whether your interest is history, marketplaces, art, or coffee, there is something for you. I have always avoided what I thought of as "traditional tours."  After experiencing RealCityTours "free" city walking tour, where you pay with a tip, it is an experience I will actively seek out on future trips. The free tour was led by a local who was an interactive storyteller. It's best to base your tours on personal recommendations and/or online reviews. I got suckered into a "transformation tour" through a popular hostel chain that was advertised as educational to learn about the positive evolution of neighborhoods. In actuality, it was a drive to three different tourist spots to look around and spend money with the tour guide’s friends. I've had a great experience and received excellent in-person feedback on the tours below. freetour.com : , , and Real City Tours Free walking tour Exotic fruits tour Transformation tour Comuna 13 Medellin Graffiti Tour Urban coffee tour Markets Marketplaces refer to anything from "mom and pop" merchants where everyday people or businesses buy groceries, to hubs with restaurants from around the world. Both are worth exploring. The former lends itself to more adventure. Traditional Marketplaces Plaza Minorista San Alejo 1st Saturday of the month San Fernando Plaza Market Square Envigado Central Mayorista de Antioquia Placita de Flórez El Pulguero Llanogrande Limited to a few times per year Restaurant Hubs Mercado del Río Resources Markets map The Best Markets in Medellín, Colombia by CultureTrip The Best Street Food Markets in Medellín by CultureTrip Social scene Through the handful of nights I went out, I experienced the variety the city offers. I explored a battle of the bands at the hostel, danced to Reggaeton till 4 AM on my birthday at , stepped on a few toes at a local salsa joint, , helped host a rooftop morning yoga and dance party, and saw a new friend DJ. Sellina Buena Vista El Tíbiri My social circle grew through roommates, informal events friends hosted, tech meetups, etc. One of my favorite experiences was . A local friend introduced me to this street where you walk up to a bodega, drink beer, and hang out on the street. People are here every night, and I hear it can get rowdy into the early morning. Calle 66a If looking for the hostel party scene, there are well-known "party hostels" like and the . For bars and clubs, the two main spots are around in El Poblado and in Laureles. Selina Purple Monkey Parque Lleras Carrera 70 An alternative to the bar scene is language exchanges like (Wednesdays), and (Wednesdays and Fridays). Locals and travelers meet in order to practice speaking, listening, and hanging out. El Social Café Ondas Outdoors Within a quick trip from downtown, you can be in the jungle jumping into a waterfall or overlooking the cityscape. The proximity of the locations and range of difficulty in accessing these sites make the samples below easy to get to from downtown. I'm happy I did last minute. I chose this because it was a decent workout, thanks to the ascent, a view of the city, and the opportunity to jump in the waterfall. The waterfall was cold, surprisingly not as cold as Yosemite National Park's waterfall in California during the summertime. Cuevas del Higuerón You can take Uber or public transportation to the trailhead at . Cuevas del Higuerón - Camino Ancestral - Camino Real La Ayurá Next time I'd love to do El Cerro De Las Tres Cruces because of how easy it is to access and the citywide view. The hike up lasts only 30-45 min. To get to the trailhead, take an Uber to which will take you to the start point. Casa de los Bernal Here are more great outdoor adventures. La Ruta hiking Facebook group Parque Arví bike Parque Arví hike Parque Arví night hike ATV tour offered by Rango Hostel Mountain bike descent offered by Rango Hostel Spas After a week of exploring, it is nice to wind down with a massage on the rooftop of the . It took me a while to book the full body massage because I had a hard time understanding what I was ordering due to my Spanish, or lack thereof. The massage was good, costing $50 per hour, about half the cost compared to the states. Diez 10 Hotel If you decide to go to Diez, the following descriptions of the menu will save you my confusion. :  Full body using hot stones Masaje terrenal :  Using coconuts Masaja tropical / nuez de coco Masaja de caña: Using bamboo : Using hot wax Masaja atardecer llanero For a less fancy and highly rated experience, is $40 per hour. Sports Massage Lastly, I explore how to work effectively, Medellín —  Work remotely If you're curious "Why Medellín?" or for the best local flavors, read the following. Medellín —The nomad's hidden gem Medellín —  A taste of classics and the best cacao enjoys sharing and is a researcher and product consultant. Adam Hurwitz travel open info

Stories, News & Mentions about CITYSCOPE

Cityscope’s Stories on HackerNoon