Modified model

[agnieszkajust]

How should I change class Decoder if I want to learn Network with additional: torch.nn.Conv2d(in_channels=16, out_channels=16, kernel_size=3, stride=1, padding=1), torch.nn.LeakyReLU(inplace=False, negative_slope=0.1) in each module in Extractor.

[Etienne66]

This is part of https://github.com/sniklaus/pytorch-pwc and you would need to not load the pre-trained model network-default.pytorch which it loads into the Flow_PWC model. Also the in_channels would be intCurrent + 128 + 128 + 96 + 64 + 32 with out_channels=16. There are several other places you would need to adjust the in_channels

It is still trainable without the pre-trained model for the flow.