two stream fusion

[yanpeilun]

Hi bryanyzhu:
thanks for your opensource code!
the spatial and temporal stream were ahieved. I want to know ,wheter the fusion of these two steram is reached ??.I didn't find the relevant source code under the project.

[bryanyzhu]

I didn't do the fusion, but I think it should reach the performance. I am working on other projects at this moment.

The source code is here, the spatial result is saved in a npy file, and the temporal result is saved in another npy file. You just need to load these two files, weighted averaging them, and get the final fusion result. It should be easy.

[Benjiou]

Hi @yanpeilun and @bryanyzhu,

1. Have you stored somewhere the ucf101_s1_rgb_resnet152.npy and ucf101_s1_flow_resnet152.npy scores. If yes do you mind sharing them ?

2. What is the final score after fusing RGB and Flow scores ?

Thank you a lot

[bryanyzhu]

Hi @Benjiou ,

1. The .npy files are generated using my provided pre-trained models. The code is also provided. You can run the code under this directory (https://github.com/bryanyzhu/two-stream-pytorch/tree/master/scripts/eval_ucf101_pytorch) to get them.

2. I don't know the fusing scores, because I didn't do it. You can try it. Thank you.