Shubham Agarwal

@aasish Could you please comment as to how should I resolve this?

FYI, I solved the issue with [this shell script](https://github.com/yahoo/FEL/blob/d6050b26cd7e6beb9ba0714e6461b392973014f6/src/main/bash/mine-datapack.sh). README needs to be updated.

@Mehrad0711 maybe you can try and integrate BS from allennlp; their implementation [here](https://github.com/allenai/allennlp/blob/master/allennlp/nn/beam_search.py)

Two suggestions to speed up the code and to avoid memory leak on other GPU: 1. Do `torch.cuda.set_device(device)` before `torch.cuda.empty_cache()` 2. `torch.cuda.empty_cache()` after every epoch instead of every batch. Had...

My hunch was that if we use parallelization in pytorch's [data loader](https://github.com/batra-mlp-lab/visdial-challenge-starter-pytorch/blob/master/train.py#L139) and still do [multiprocess tokenization](https://github.com/batra-mlp-lab/visdial-challenge-starter-pytorch/blob/master/visdialch/data/readers.py#L129) in one process, it is giving those errors. Basically, it tries to do...

Yeah I did try with 1 worker. Had the same errors. (Cant use 0 because [this](https://github.com/batra-mlp-lab/visdial-challenge-starter-pytorch/blob/master/visdialch/data/readers.py#L129) requires at least one worker :D ) Have removed multiprocess tokenization in my code...

Thanks for the prompt reply. :)

@abhinav3 Do you have an open PR for this?

@woozzu What do you suggest should be an ideal way to take images of size 256/224? Should we just reshape the image to 128 or is the code generalized enough...

Hi, Is there any update for this? I know FB's [visdom](https://github.com/facebookresearch/visdom) repo could be a starting point. However, in my initial experiments with it, I didn't have aligned images and...