Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

What if cows could have their own pose estimation model? 🐄 Cow pose estimation using Ultralytics YOLO26 Pose, with 20 keypoints trained on a custom dataset. The dataset is also available for testing directly in the Platform. Link 👇 Computer vision keeps getting more interesting. 👀 #Cow #Pose #MachineLearning

Muhammad Rizwan Munawar

2,190 subscribers

21,264 views • 15 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

17 Comments

Muhammad Rizwan Munawar15 days ago

Dataset:

Rubén14 days ago

@ultralytics Holy cow!

Muhammad Rizwan Munawar14 days ago

@ultralytics Haha, yes ;(

fr34kΞth14 days ago

@ultralytics I have a training set for you

Muhammad Rizwan Munawar14 days ago

@ultralytics Wow, nice, I would love to explore :)

Ultralytics15 days ago

Wow, nice pose 🐄

Muhammad Rizwan Munawar15 days ago

Thank you 😀

NachoAM14 days ago

@ultralytics Hi Muhammad, great work on the cow pose model! I tested it and it looks promising. Could you share the names/order of its 20 keypoints (and the skeleton connections, if available)? Thanks!

Muhammad Rizwan Munawar14 days ago

@ultralytics Thanks alot :) Yes, the kpts connections and names can be found in the code available here:

NachoAM14 days ago

@ultralytics Thank you so much! This is exactly what I needed. I really appreciate you sharing the script and the skeleton layout.

Muhammad Rizwan Munawar14 days ago

@ultralytics My pleasure :)

Vexl14 days ago

@ultralytics VEXL simplifies smart device prototyping

Muhammad Rizwan Munawar14 days ago

@ultralytics Wow, nice, I would love to explore!

Vexl14 days ago

@ultralytics That's great, but for now you can join our beta

frygorate14 days ago

@ultralytics kinda new to computer vision but why do some of the skeletons keep disappearing or become wildly inaccurate for a split second its kinda pissing me off? is there a fix for that?

Muhammad Rizwan Munawar14 days ago

@ultralytics I think more annotations could improve that :)

Rahul Deora14 days ago

@ultralytics all this is meaningless. What is required is action recognition and good event description of long videos. Shipping these basic features not going to do much @muhammdrizwanmr

Related Videos

Vehicle counting using Ultralytics YOLO26 🚀 Here, I trained the YOLO26 model on a custom dataset and later used an object tracker (Bytetrack) + Ultralytics Solutions to count the objects in the video file. #vehicles #tracking #MachineLearning

Vehicle counting using Ultralytics YOLO26 🚀 Here, I trained the YOLO26 model on a custom dataset and later used an object tracker (Bytetrack) + Ultralytics Solutions to count the objects in the video file. #vehicles #tracking #MachineLearning

Muhammad Rizwan Munawar

46,738 views • 3 months ago

hi all, excited to join! i'm building an expressive mini "shoggoth" robot which will eventually be hooked up to gpt4o realtime voice. i'm currently working on the low-level policies, which are trained in a mujoco simulation with RL. to delay working on raw-pixels for now, i trained a pose-estimation model using deeplabcut and triangulate the position in 3d space using the stereo cameras. eventually, i'll use gpt4o's tool calling capabilities to activate several of these policies (closed and open loop) based on the dialog flow! captions: manual actuation of the tentacle / 3d pose estimation / target design

hi all, excited to join! i'm building an expressive mini "shoggoth" robot which will eventually be hooked up to gpt4o realtime voice. i'm currently working on the low-level policies, which are trained in a mujoco simulation with RL. to delay working on raw-pixels for now, i trained a pose-estimation model using deeplabcut and triangulate the position in 3d space using the stereo cameras. eventually, i'll use gpt4o's tool calling capabilities to activate several of these policies (closed and open loop) based on the dialog flow! captions: manual actuation of the tentacle / 3d pose estimation / target design

Matthieu LC

45,587 views • 1 year ago

Workouts monitoring using Ultralytics YOLO26 pose 🏃‍♂️‍➡️ Imagine tracking your workouts using YOLO26, whether it's leg press, bench press, or leg extension, with just 2 lines of code! I've shared the command for running it, making it easy to track your workouts even on a CPU-based system. You don’t always need a high-end GPU; exporting the model to TorchScript or ONNX can boost performance on CPU devices. Code below 👇 #Workout #AI #MachineLearning

Workouts monitoring using Ultralytics YOLO26 pose 🏃‍♂️‍➡️ Imagine tracking your workouts using YOLO26, whether it's leg press, bench press, or leg extension, with just 2 lines of code! I've shared the command for running it, making it easy to track your workouts even on a CPU-based system. You don’t always need a high-end GPU; exporting the model to TorchScript or ONNX can boost performance on CPU devices. Code below 👇 #Workout #AI #MachineLearning

Muhammad Rizwan Munawar

15,723 views • 13 days ago

The BEDLAM2.0 dataset (B2) is here, just in time to train your 3D human pose and shape estimation methods for CVPR. B2 goes beyond BEDLAM (B1) to include widely varied and natural camera motions and fields of view, more diverse body shapes, strand-based hair, more garments, shoes, more body motions, and more 3D scenes. Compared with B1, training on B2 produces more accurate 3D human pose, resulting in SOTA accuracy, particularly for estimates in world coordinates. B2 lets you jointly train camera motion and human motion regressors, and we also provide depth maps. Check out data, code, dataset statistics, and much more. BEDLAM2.0 will appear in the 2025 NeurIPS Datasets and Benchmarks Track. Joint work with Joachim Tesch, Giorgio Becherini, Prerana Achar, Anastasios Yiannakidis, Muhammed Kocabas, Priyanka Patel.

The BEDLAM2.0 dataset (B2) is here, just in time to train your 3D human pose and shape estimation methods for CVPR. B2 goes beyond BEDLAM (B1) to include widely varied and natural camera motions and fields of view, more diverse body shapes, strand-based hair, more garments, shoes, more body motions, and more 3D scenes. Compared with B1, training on B2 produces more accurate 3D human pose, resulting in SOTA accuracy, particularly for estimates in world coordinates. B2 lets you jointly train camera motion and human motion regressors, and we also provide depth maps. Check out data, code, dataset statistics, and much more. BEDLAM2.0 will appear in the 2025 NeurIPS Datasets and Benchmarks Track. Joint work with Joachim Tesch, Giorgio Becherini, Prerana Achar, Anastasios Yiannakidis, Muhammed Kocabas, Priyanka Patel.

Michael Black

27,667 views • 10 months ago

For all my lifters: computer vision app to measure back curvature during deadlift! main technical highlights: — RF-DETR (Roboflow) to segment the person (great performance out-the-box with no additional training!) — YOLO11n (Ultralytics) for bounding box prediction around the barbell weight plate (trained on my own small dataset). — Mediapipe (Google) for pose landmark detection to guide the bounds for the line fitting. — Custom logic to fit a line across the back. about the plots on the right side: The Back Roundness Map shows the deviation of the estimated back curve from a line of best fit. Credit: Jeremy Park —— if it matters in AI or Robotics you'll see it here first:

For all my lifters: computer vision app to measure back curvature during deadlift! main technical highlights: — RF-DETR (Roboflow) to segment the person (great performance out-the-box with no additional training!) — YOLO11n (Ultralytics) for bounding box prediction around the barbell weight plate (trained on my own small dataset). — Mediapipe (Google) for pose landmark detection to guide the bounds for the line fitting. — Custom logic to fit a line across the back. about the plots on the right side: The Back Roundness Map shows the deviation of the estimated back curve from a line of best fit. Credit: Jeremy Park —— if it matters in AI or Robotics you'll see it here first:

Ilir Aliu

234,710 views • 7 months ago

It's way too many people commenting on black culture that have 0 knowledge on its origins. Kendrick is doing what we call a jail pose. Has kendrick been to jail? No but he survived the trapings of the system. Also one could argue that the pose comes from us taking pictures with a lot of people so some have to squat to be in the picture. Ultimately its a hood thing.

It's way too many people commenting on black culture that have 0 knowledge on its origins. Kendrick is doing what we call a jail pose. Has kendrick been to jail? No but he survived the trapings of the system. Also one could argue that the pose comes from us taking pictures with a lot of people so some have to squat to be in the picture. Ultimately its a hood thing.

Reed

119,400 views • 6 months ago

Meta presents Sapiens Foundation for Human Vision Models discuss: We present Sapiens, a family of models for four fundamental human-centric vision tasks - 2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction. Our models natively support 1K high-resolution inference and are extremely easy to adapt for individual tasks by simply fine-tuning models pretrained on over 300 million in-the-wild human images. We observe that, given the same computational budget, self-supervised pretraining on a curated dataset of human images significantly boosts the performance for a diverse set of human-centric tasks. The resulting models exhibit remarkable generalization to in-the-wild data, even when labeled data is scarce or entirely synthetic. Our simple model design also brings scalability - model performance across tasks improves as we scale the number of parameters from 0.3 to 2 billion. Sapiens consistently surpasses existing baselines across various human-centric benchmarks. We achieve significant improvements over the prior state-of-the-art on Humans-5K (pose) by 7.6 mAP, Humans-2K (part-seg) by 17.1 mIoU, Hi4D (depth) by 22.4% relative RMSE, and THuman2 (normal) by 53.5% relative angular error.

Meta presents Sapiens Foundation for Human Vision Models discuss: We present Sapiens, a family of models for four fundamental human-centric vision tasks - 2D pose estimation, body-part segmentation, depth estimation, and surface normal prediction. Our models natively support 1K high-resolution inference and are extremely easy to adapt for individual tasks by simply fine-tuning models pretrained on over 300 million in-the-wild human images. We observe that, given the same computational budget, self-supervised pretraining on a curated dataset of human images significantly boosts the performance for a diverse set of human-centric tasks. The resulting models exhibit remarkable generalization to in-the-wild data, even when labeled data is scarce or entirely synthetic. Our simple model design also brings scalability - model performance across tasks improves as we scale the number of parameters from 0.3 to 2 billion. Sapiens consistently surpasses existing baselines across various human-centric benchmarks. We achieve significant improvements over the prior state-of-the-art on Humans-5K (pose) by 7.6 mAP, Humans-2K (part-seg) by 17.1 mIoU, Hi4D (depth) by 22.4% relative RMSE, and THuman2 (normal) by 53.5% relative angular error.

AK

151,614 views • 2 years ago

ViPE: Video Pose Engine for 3D Geometric Perception Contributions: • A robust and efficient framework, ViPE, for estimating camera parameters and dense depth from diverse, in-the-wild videos. • A system design that integrates the strengths of classical SLAM (efficiency, scalability) and learned models (robustness), with key improvements in efficiency, dynamic object handling, and depth quality over prior work. • A large-scale dataset of annotated videos, created using ViPE, to facilitate future research in 3D computer vision.

ViPE: Video Pose Engine for 3D Geometric Perception Contributions: • A robust and efficient framework, ViPE, for estimating camera parameters and dense depth from diverse, in-the-wild videos. • A system design that integrates the strengths of classical SLAM (efficiency, scalability) and learned models (robustness), with key improvements in efficiency, dynamic object handling, and depth quality over prior work. • A large-scale dataset of annotated videos, created using ViPE, to facilitate future research in 3D computer vision.

MrNeRF

42,644 views • 1 year ago

// The Harmony of Pose-to-Pose and Layered Methods Explained (Continued) 1. In my previous post, I demonstrated the evolution of a standing roundhouse kick starting with Pose-to-Pose and transitioning into the Layered Method. 2. for this Hulk animation, I've opted for a different presentation format to ensure absolute clarity in showcasing the process. 3. the first 4 steps are how I animate 99.99999% of the time whether I'm working on AAA Games, Films, mobile or TV shows // Step #1 - I pose my character using 5 key poses (1 & 5 are duplicates. // Step #2 - I set the timing by moving ALL KEYS on the timeline until my eyes see something that works // Step #3 - I start adjusting my timing & F-curves. // Step #4 - I start using the Layered Method drop any questions you have in the comments 👇, I will answer them when I have a moment and a ☕️ in hand! ;)

// The Harmony of Pose-to-Pose and Layered Methods Explained (Continued) 1. In my previous post, I demonstrated the evolution of a standing roundhouse kick starting with Pose-to-Pose and transitioning into the Layered Method. 2. for this Hulk animation, I've opted for a different presentation format to ensure absolute clarity in showcasing the process. 3. the first 4 steps are how I animate 99.99999% of the time whether I'm working on AAA Games, Films, mobile or TV shows // Step #1 - I pose my character using 5 key poses (1 & 5 are duplicates. // Step #2 - I set the timing by moving ALL KEYS on the timeline until my eyes see something that works // Step #3 - I start adjusting my timing & F-curves. // Step #4 - I start using the Layered Method drop any questions you have in the comments 👇, I will answer them when I have a moment and a ☕️ in hand! ;)

John Animates

18,873 views • 3 years ago

🚀 Struggling with the lack of high-quality data for AI-driven human-object interaction research? We've got you covered! Introducing HUMOTO, a groundbreaking 4D dataset for human-object interaction, developed with a combination of wearable motion capture, SOTA 6D pose estimation vision models, LLM, and the professional refining works of multiple animation studios. HUMOTO features: ✅ Over 700 diverse daily activities ✅ Interactions with 60+ objects, 70+ articulated parts. ✅ Fine-grained text annotations ✅ Detailed hand and finger movements We hope HUMOTO will fuel your Humanoid AI research and drive new advancements! For research or commercial license inquiries, please contact yizho@adobe.com. Explore the dataset: 👉 HUMOTO Dataset Website Learn more: 👉 HUMOTO Project Page Jiaxin Lu Qixing Huang #AI #HumanObjectInteraction #HumanoidAI #MotionCapture #HUMOTO #AdobeResearch

🚀 Struggling with the lack of high-quality data for AI-driven human-object interaction research? We've got you covered! Introducing HUMOTO, a groundbreaking 4D dataset for human-object interaction, developed with a combination of wearable motion capture, SOTA 6D pose estimation vision models, LLM, and the professional refining works of multiple animation studios. HUMOTO features: ✅ Over 700 diverse daily activities ✅ Interactions with 60+ objects, 70+ articulated parts. ✅ Fine-grained text annotations ✅ Detailed hand and finger movements We hope HUMOTO will fuel your Humanoid AI research and drive new advancements! For research or commercial license inquiries, please contact [email protected]. Explore the dataset: 👉 HUMOTO Dataset Website Learn more: 👉 HUMOTO Project Page Jiaxin Lu Qixing Huang #AI #HumanObjectInteraction #HumanoidAI #MotionCapture #HUMOTO #AdobeResearch

Yi Zhou

71,609 views • 1 year ago

People are using Verdent to build projects that are both interesting and genuinely useful. Some are already using computer vision to empower the sports industry. Take this badminton video analysis demo. The result is not perfect: shuttle tracking jitters, speed is only estimated, and pose detection still needs real domain tuning. But in just one conversation, Verdent turned a fuzzy idea into a playable 25-second annotated video: player boxes, body keypoints, shuttle markers, a speed HUD, and an exported MP4.

People are using Verdent to build projects that are both interesting and genuinely useful. Some are already using computer vision to empower the sports industry. Take this badminton video analysis demo. The result is not perfect: shuttle tracking jitters, speed is only estimated, and pose detection still needs real domain tuning. But in just one conversation, Verdent turned a fuzzy idea into a playable 25-second annotated video: player boxes, body keypoints, shuttle markers, a speed HUD, and an exported MP4.

Verdent

13,040 views • 3 months ago

"We used to have some HEAVY HITTERS... THIS IS A F***IN" PLUMBER LEAGUE‼️" Aches SOUNDS OFF on the praise for Shotzzy, claiming today's competition doesn't pose a strong enough challenge- 👇 Agree or disagree with the SLAYA? 👇

"We used to have some HEAVY HITTERS... THIS IS A F***IN" PLUMBER LEAGUE‼️" Aches SOUNDS OFF on the praise for Shotzzy, claiming today's competition doesn't pose a strong enough challenge- 👇 Agree or disagree with the SLAYA? 👇

The Flank

170,877 views • 6 months ago

The largest fully annotated open egocentric human dataset. Today we're open-sourcing EgoSuite-Open100K with Hugging Face : 100,000 hours of egocentric human data, free for anyone to use. Physical AI has a scaling law of its own now, and its input is human data. That part is settled. What the field still needs is a shared foundation to build on. 15,000+ tasks. 15,000+ real scenes, from factory floors to retail backrooms. Hand pose, body pose and subtask-level semantics, with wrist camera coverage on part of the set. Licensed for commercial training, not research-only. We build the data infrastructure for Physical AI. Open100K is the first public layer of it. The first 10,000 hours are live now, with the rest rolling out in stages.

The largest fully annotated open egocentric human dataset. Today we're open-sourcing EgoSuite-Open100K with Hugging Face : 100,000 hours of egocentric human data, free for anyone to use. Physical AI has a scaling law of its own now, and its input is human data. That part is settled. What the field still needs is a shared foundation to build on. 15,000+ tasks. 15,000+ real scenes, from factory floors to retail backrooms. Hand pose, body pose and subtask-level semantics, with wrist camera coverage on part of the set. Licensed for commercial training, not research-only. We build the data infrastructure for Physical AI. Open100K is the first public layer of it. The first 10,000 hours are live now, with the rest rolling out in stages.

Lightwheel

73,911 views • 28 days ago

Character LoRA on Flux – sneak peak 👀 I trained a model on 15 character images (same pose, different expressions) and now I can generate endless variants in new setups and poses. One image per inference, perfect consistency (both style + subject) with over 90% usable straight out of the box. Video below 👇 Trained on Scenario today.

Character LoRA on Flux – sneak peak 👀 I trained a model on 15 character images (same pose, different expressions) and now I can generate endless variants in new setups and poses. One image per inference, perfect consistency (both style + subject) with over 90% usable straight out of the box. Video below 👇 Trained on Scenario today.

Emm | scenario.com

48,818 views • 1 year ago

#JOSHUA #조슈아 weverse live Shua's turn to corner Jeonghan today🤭 and ends up asking for recommendations 🦌: they are curious about our unit pose Jeonghan-ah. Wasn't there one which you created previously. What was it? It's hazy for me. There was that pose which you said we definitely have to go with 😇: Ah but that one that was a different one 🦌: no you.. no! for the unit you.. 😇: That wasn't the unit pose. You asked me what pose should we show to carats and I asked how about this 🦌: what are you talking about. I can hear the gears turning in your head so clearly 😇: No I was thinking 'did I do a unit pose' 🦌: you did a unit pose that time 😇: that was a pose which you wanted to look sexy 🦌: No! 😇: show us that 🦌: you were like 'we definitely have to do this. Carats will definitely..' 😇: that if you are sexy carats will definitely like it? 🦌: No Jeonghan-ah quickly show it. Ayy 😇: what is it? Tell me. Even if it is by whispering it to me 🦌: I don't remember it exactly 😇: It's because I don't remember it 🦌: There is that! The one that you showed me 😇: I don't know. My memory isn't that good 🦌: Ah your memory isn't good? 😇: Amnesia 🦌: Please give us recommendations

#JOSHUA #조슈아 weverse live Shua's turn to corner Jeonghan today🤭 and ends up asking for recommendations 🦌: they are curious about our unit pose Jeonghan-ah. Wasn't there one which you created previously. What was it? It's hazy for me. There was that pose which you said we definitely have to go with 😇: Ah but that one that was a different one 🦌: no you.. no! for the unit you.. 😇: That wasn't the unit pose. You asked me what pose should we show to carats and I asked how about this 🦌: what are you talking about. I can hear the gears turning in your head so clearly 😇: No I was thinking 'did I do a unit pose' 🦌: you did a unit pose that time 😇: that was a pose which you wanted to look sexy 🦌: No! 😇: show us that 🦌: you were like 'we definitely have to do this. Carats will definitely..' 😇: that if you are sexy carats will definitely like it? 🦌: No Jeonghan-ah quickly show it. Ayy 😇: what is it? Tell me. Even if it is by whispering it to me 🦌: I don't remember it exactly 😇: It's because I don't remember it 🦌: There is that! The one that you showed me 😇: I don't know. My memory isn't that good 🦌: Ah your memory isn't good? 😇: Amnesia 🦌: Please give us recommendations

🦌💎

20,422 views • 1 month ago

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. There are 4 key contributions that make it so accurate and robust: 1. To get accurate 3D shape and pose as well as good alignment to image features, you need to know the focal length of the camera. To solve this, we train HumanFOV to compute the field of view. 2. We introduce CameraHMR, which integrates HumanFOV into HMR2.0 to exploit the estimated focal length. 3. To get accurate pseudo ground truth (pGT) training data, we compute the focal length for images in 4DHumans dataset and modify SMPLify to take this into account. 4. But SMPLify only uses sparse 2D keypoints, which do not capture body shape. So we train a dense surface keypoint detector, DenseKP, on BEDLAM and run it on 4DHumans, resulting in improved body shape. The resulting method is CamSMPLify. We iterate training CameraHMR and running CamSMPLify on the training set initialized with CameraHMR. This results in much improved pGT for 4DHumans and a SOTA single-image HMR method.

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. There are 4 key contributions that make it so accurate and robust: 1. To get accurate 3D shape and pose as well as good alignment to image features, you need to know the focal length of the camera. To solve this, we train HumanFOV to compute the field of view. 2. We introduce CameraHMR, which integrates HumanFOV into HMR2.0 to exploit the estimated focal length. 3. To get accurate pseudo ground truth (pGT) training data, we compute the focal length for images in 4DHumans dataset and modify SMPLify to take this into account. 4. But SMPLify only uses sparse 2D keypoints, which do not capture body shape. So we train a dense surface keypoint detector, DenseKP, on BEDLAM and run it on 4DHumans, resulting in improved body shape. The resulting method is CamSMPLify. We iterate training CameraHMR and running CamSMPLify on the training set initialized with CameraHMR. This results in much improved pGT for 4DHumans and a SOTA single-image HMR method.

Michael Black

21,696 views • 1 year ago