search

Hrener / 3D-Action-recognition

PyTorch implementation of Two-stream CNN for 3D action recognition
149 stars 18 forks source link

What data modalities used in the two streams? #1

Open liangchengwu opened 5 years ago

liangchengwu commented 5 years ago

  1. What are the input data modalities for two streams? Depth maps and Skeleton? or RGB and optical flow?

  2. For the output of two streams, did you do the fusion at a fully connected layer and then fed them to softmax for label prediction?

Thanks！

Hrener commented 5 years ago

Sorry, the reply is too late. The input is the 3D skeleton data of NTU RGB+D, and the data of the two streams is spliced and sent to the fully connected layer after the feature extraction.