search

Diananini / LD-ConGR-CVPR2022

A Large RGB-D Video Dataset for Long-Distance Continuous Gesture Recognition
23 stars 2 forks source link

Is there a light-model version? #3

Closed erwangccc closed 2 years ago

erwangccc commented 2 years ago

Maybe current model is unfriendly to edge devices, is there a light-model version? Thanks

Diananini commented 2 years ago

The model in this paper is not designed for edge devices. Maybe we will try to propose a lightweight gesture recognition model in the future.

erwangccc commented 2 years ago

Thanks for your quick response. Looking forward to your future great work. Do you have any recommended light gesture recognition models based on RGB-D data? Thanks in advance.

Diananini commented 2 years ago

As far as I know, there is no lightweight model specifically for gesture recognition. Maybe you can try lightweight models proposed for action recognition, such as X3D.

erwangccc commented 2 years ago

Thank you. I want to confirm that, the paper says the gesture(hand) region is predicted and it will enlarge the hand region, so you will resize all 32/16 frames to input size 112x112, right? But we do resize directly, the hand will deform, is it OK?

Diananini commented 2 years ago

Resize is used in both training and testing, and there is deformation in both training and testing. The key is that the distribution of test data is consistent with the training data, rather than one with deformation and the other without.

erwangccc commented 2 years ago

Thank you. Maybe padding by zero is ok,right?

Diananini commented 2 years ago

Yes, you can try it.