Rinat M

Update requirements.txt

2

Since in PyTorch 2 dashes are preferred over underscores (https://github.com/pytorch/pytorch/pull/94505), **local-rank** that is propagated by distributed training `torch.distributed.launch` script, it is not compatible with the python training scripts. This commit...

Improves compatibility with PyTorch 2, especially with torch.distributed.run or torchrun scripts.

It seems to work well with the pytorch 1.8.0 and 1.8.1 and cuda 11.1. It allows to install the correct dependencies on Ubuntu 18 and 20 with the proposed _pip...

I think since the '_summarize' function expects a _maxDets_ argument and a user is allowed to specify _params.maxDets_, it is crucial to pass it; otherwise, it is going to always...

To run a test **dataset_dir** is not required . Otherwise, **test.py** just doesn't work since it is trying to check for a _raw_ subfolder in the _datasets_ folder.