Closed shuzhaolei closed 5 years ago
And I am trying to re-implement your work with PyTorch. I don't know coding differences between PyTorch and Caffe2 exactly.
It is a design choice, in later model we use MaxPool but with higher resolution (https://arxiv.org/abs/1904.02811). Code & models are released soon.
Why there is no maxpool layer after Conv1? Based on ResNet, I think it should be a maxpool layer around R2Plus1D/lib/models/r3d_model.py line179. But in your code, I couldn't find one.
Waiting for your answer, thx!