speed and pytorch version about two-stream-pytorch HOT 4 CLOSED

For the pytorch version, it should be 0.2 or 0.3 because this is a project developed 2 years ago. But if you want to work with the latest pytorch version, it should be fine or requiring very minimum of change because this repo doesn't have any customized layer or operation.

Since this is a standard classification problem, the speed is just the network's speed. I mean, if you know resnet50's speed, our model should have the same speed.

Other requirements may be just opencv.

from two-stream-pytorch.

can we use the python-opencv, not the make version from source code?

from two-stream-pytorch.

If you want to extract optical flow, you need build from source.

But if you only want to train the model, no matter spatial stream or temporal stream, python-opencv is enough.

from two-stream-pytorch.

ok, thank you!

from two-stream-pytorch.