Features extraction at frame level

[pinkfloyd06]

Hello @shuangshuangguo

Let me first thank you for the conversion.

Can l extract features representation (from last fully connected layer) for each frame (rather than for the whole video ) for the RGB stream and optical flow stream ?

Thank you a lot

[shuangshuangguo]

Yes, you can do that absolutely by modifying code. Specifically, you should modify dataset.py to input one frame each time.

[Benjiou]

Thank you for your answer @shuangshuangguo, but l can't find dataset.py in https://github.com/shuangshuangguo/caffe2pytorch-tsn

Thank you

[shuangshuangguo]

This repo just converse caffe model to pytorch model, thus does not contain the TSN source code. You can find dataset.py in https://github.com/yjxiong/tsn-pytorch.

[770258506]

@shuangshuangguo Thank you for your answer，where can i find the code to extract features?