search

jeffreyyihuang / two-stream-action-recognition

Using two stream architecture to implement a classic action recognition method on UCF101 dataset
MIT License
858 stars 249 forks source link

choice of making flow tensor #48

Open ratnesh1729 opened 5 years ago

ratnesh1729 commented 5 years ago

https://github.com/jeffreyhuang1/two-stream-action-recognition/blob/master/dataloader/motion_dataloader.py#L55 variable j starts from 0, so according to line 55 and 56, first frames optical flow - u and v - will be in: flow[-2,:,:] and flow[-1,:,:]. While it makes sense for it to be flow[0,:,:] and flow[1,:,:] ?

ratnesh1729 commented 5 years ago

https://github.com/jeffreyhuang1/two-stream-action-recognition/issues/28 addresses this to an extent - but the code is still incorrect.