Official YOLOv7 Pose vs MediaPipe | Full comparison of real-time Pose Estimation | Which is Faster? 

R Ttf P P

Glad you liked it @John. More videos incoming!

Awesome, thank you!

We felt the same while working with both YOLOv7 and mediapipe that everyone should know about this comparison! Glad you found it useful.

You would need to retrain the network with a different backbone. The authors have trained it for the YOLOv7-W6 model. You can train the model using a different yolov7 model. What you would need is a config (.yaml) file corresponding to the smaller model. You can then train the model using the commands given here: github.com/WongKinYiu/yolov7/tree/pose I doubt it would give accurate results for smaller models. I would use mediapipe if I don't need multi-person pose estimation.

As of 2024 Jan update, Mediapipe does supports mutiperson pose but limited to 5 at a time. For further info check out: developers.google.com/mediapipe/solutions/vision/pose_landmarker/

Thanks for the kind words Geoff! YOLO does not have good enough number of points for Face landmarks alignment. Mediapipe has a dedicated face mesh model that gives 468 3D landmark points on the face. You can check out our blog post on Creating Snapchat filters using mediapipe. You can learn about how to use the different points for your application. learnopencv.com/create-snapchat-instagram-filters-using-mediapipe/

Glad it was helpful!

Glad it is helpful!

Thank you, glad you liked it 😊

Thank You!

Thank you Yohanes!

Thank you so much!

Thank you. Glad you liked it.

As mentioned in the summary section, it's better to use YOLOv7 or other pose models as mediapipe is optimized for real-time performance which is more suitable for video inference. Hope that helps!

@@LearnOpenCV so with Multiperson which is better than yoloV7.

@@dj.qb91 For Multiperson we're checking out MMPose next -> github.com/open-mmlab/mmpose. You may also check it out and compare with YOLOv7. Check this out for getting started: mmpose.readthedocs.io/en/v0.29.0/get_started.html#inference-with-pre-trained-models

@@LearnOpenCV thanks 🙏🏾

The pose solution model consists of two models. The detection model (that detects the body), and the landmark model (that maps the landmarks). If you can make the detection model detect the body without its upper part, theoretically, the solution will work.

I'm not sure what that means, but I'm hoping you liked it! 😊

You are very welcome

Hi Maxim Are you talking about face Detection or Facial Landmarks Detection using YOLOv7?

Hey! I’m talking about Facial Landmarks Detection. I fine-tuned and used ensemble instead

Great, do you have a repo you could share?

@@LearnOpenCV I worked with medical sensitive data(

No Issues!