How to train the network.

[Serenos]

Thanks for releasing your code! Can you show more about how to train the network?

[Sixkplus]

Thank you for your raising the problem. As mentioned in the paper, TapLab focuses on the design of acceleration modules, which is independent from the training process. For the training part, you can train the per-frame model using any exsiting script(e.g. TorchSeg), and then load the trained model in the inference script.