mchaniotakis

Maybe the mmcv version is to old?

There is something wrong with your environment. I suggest taking a day to understand all of the dependencies -> mmengine , mmcv , pytorch + cuda , cuda with (pplcv...

Great work! Could we also include queryInst in this PR since they are very similar?

If possible, I would try to split the grid_sample into multiple objects within the python code if I were in your shoes.

I observe the same issue across all DV movies I have watched, if there was a setting to raise the blacks (like there is for brightness) I could manually adjust...

Just speculating here, the second GPU seems almost at capacity, i have had memory issues that froze training. Have you tried training with fewer samples past 300? I would say...

Wouldn't just a crop augmentation achieve what you need for the training step?