search

facebookresearch / InterHand2.6M

Official PyTorch implementation of "InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image", ECCV 2020
Other
675 stars 92 forks source link

Visualising in real time through webcam #121

Open toxicrecker opened 2 years ago

toxicrecker commented 2 years ago

Pretty self explanatory, how can i use the webcam data to track and visualise the frames in real time?

mks0601 commented 2 years ago

Webcam demo is not available, but you can hack the demo codes to make it work with the webcam

toxicrecker commented 2 years ago

the problem is that image demo code needs a bbox to function, if i change the bbox to fit the whole image it doesn't functions very well, so can you please guide me a bit on how to dynamically get the bbox of hands from the frames

mks0601 commented 2 years ago

That is true. One of my ongoing work tackles such problem.

toxicrecker commented 2 years ago

okay thank you for the information, please do let me know in this thread when it is done thank you

toxicrecker commented 1 year ago

Hello, good sir. Is there any progress on this?

toxicrecker commented 1 year ago

Is this still being worked on?

wefwwef4 commented 1 year ago

Now I use mediapipe to make bbx, but how can I make it on real-time cam?

mks0601 commented 1 year ago

Please use this https://github.com/facebookresearch/InterWild

wefwwef4 commented 1 year ago

https://github.com/facebookresearch/InterHand2.6M/assets/68526524/d7707a5c-8b4f-4ecd-9f8a-121f2ffb7358

I ‘ve make the processed_image into the full video,but i found the result is not that good, I wonder why I cant get that good result? I 'll appreciate it if you can answer the question

mks0601 commented 1 year ago

Please use this https://github.com/facebookresearch/InterWild

InterHand dataset provides 3D GT, which is good, but one limitation is that its image appearance is far from that of real in-the-wild images. For more details, please read InterWild paper.