Very nice work! But maybe there is an oversight in main_stereo.py. After validation in training stage, model.training is still False since model.train() is not called until the new epoch is coming. When the length of train_loader is not equal with the val_freq, this may result in losing several supervisions of disparity except the last one.
Very nice work! But maybe there is an oversight in
main_stereo.py. After validation in training stage,model.trainingis stillFalsesincemodel.train()is not called until the new epoch is coming. When the length oftrain_loaderis not equal with theval_freq, this may result in losing several supervisions of disparity except the last one.