YapengTian
commented
4 years ago Haha~ Yes, the order: "h, w" is better. But, it will not affect results. In addition, you could use more gpus for fast training and fine-tuning can improve performance (for example, we can first train the model from lr=1e-4. After 600 epochs, we can fine-tune the model with lr=5e-5. It is pretty similar to training strategy in EDVR.)
YoungJoongUNC
Hello!
In model.py line 219 - 223
batch_size, num, ch, w, h = x.size() # 5 video frames# center frame interpolationcenter = num // 2# extract featuresy = x.view(-1, ch, w, h)Shouldn't the order of w and h be changed as follows?
batch_size, num, ch, h, w = x.size() # 5 video frames# center frame interpolationcenter = num // 2# extract featuresy = x.view(-1, ch, h, w)