search

JacobYuan7 / DIN-Group-Activity-Recognition-Benchmark

[ICCV 2021] A new codebase containing various methods for Group Activity Recognition. Paper title: Spatio-Temporal Dynamic Inference Network for Group Activity Recognition.
MIT License
53 stars 15 forks source link

Optical flow code #21

Closed 199943 closed 1 year ago

199943 commented 1 year ago

Hello author, thank you for your wonderful work. Can you tell me how to process two channels of optical flow images using Resnet-18 and where you got the pre-trained model from, I didn't find the right code, can you provide this part of the code? thanks.

JacobYuan7 commented 1 year ago

@199943 Hi, I do not use optical flows in this work. However, if you want to use optical flows, you can transform it to 3-channel representations. I think this transformation is rather easy if you want to visualize optical flows. Then, you can adopt ResNet-18.

199943 commented 1 year ago

Thanks for the clarification.

JacobYuan7 commented 1 year ago

@199943 Note that if you transform it to three-channel representations, the ImageNet pre-training for ResNet-18 might not be helpful. This might require further study.

199943 commented 1 year ago

OK, thank you very much!