ReActNet architecture

[CosimoRulli]

Hello developers, Thank you for sharing your code. Would you mind sharing also the full instructions to reproduce your results with the training set from ReActNet? In particular, did you binarize just the activations during stage 1 or did you train the model in full precision?

[z-hXu]

We followed exactly the same training strategy as ReActNet, you can find more details at https://github.com/liuzechun/ReActNet/tree/master/resnet. Specifically, (1) we only binarize activations during stage 1, (2) we didn't train the model in full precision from scratch, instead, we download the full precision model provided by torchvision.models.