search

kenshohara / 3D-ResNets-PyTorch

3D ResNets for Action Recognition (CVPR 2018)
MIT License
3.91k stars 933 forks source link

Only input one clip for training during training？ #217

Open slighting666 opened 4 years ago

slighting666 commented 4 years ago

When training, only one clip is input for training. How to set a video to be divided into multiple clips for training, and finally merge into a video-level score?

guilhermesurek commented 4 years ago

I don't think there is this capability, so you need to custom it. I think the better way would be in the moment the code is setting up the data, functions get_training_data, get_validation_data, get_inference_data. Furthermore, you will need to not apply shuffle selection of the frames in the clip during training or validation.