erlemar/wheat

Wheat Detection challenge on Kaggle

wheat's Issues

Why is there no losses.backward and optimizer.step on training_steps?

I am using this as a baseline on my custom dataset. While I have used Pytorch for FasterRCNN before (not with Pytorch Lightning), I am curious as to why it does not have the typical ending of each training step like the ff:

...
losses = sum(loss for loss in loss_dict.values())
optimizer.zero_grad()
losses.backward()
optimizer.step()
...

I am not seeing this being implemented explicitly in the code and the utils. Am I missing something why isnt this included?

Pip install -r requirements.txt fails

Hi there I've tried to pip install the dependencies but I get the following error both on Ubuntu 18 and macOS ERROR: Could not find a version that satisfies the requirement pytorch_lightning==0.8.0. Any idea? Did you use python2?