Question: Videos do not seem to load correctly

[walsharry]

Dear Authors,

I have been trying to use the code by running "python main.py --asl_dataset msasl --checkpoint checkpoint/msasl_i3d_pbsl1k --datasetname msasl --init_cross_language asl_with_bsl --num-classes 1000 --num_gpus 1 --num_in_frames 64 --pretrained misc/pretrained_models/model.pth.tar --ram_data 1 --workers 16" but the data["rgb"] tensor seems to be empty. The output is also showing up blank.

This seem to happen on both the WLASL and MSASL datasets. The folder structure is as follows.
data
|-experiments
| ---wlasl_i3d_pbsl1k
|-msasl
| ---info
| ---videos_original
|-wlasl
| ---info
| ---videos_original
| -----test
| -----train
| -----val

What can I do to fix this issue? I understand this is most likely issue with my setup, but any help is very much appreciated. Thanks in advance.

[gulvarol]

Could you try with --ram_data 0 instead, i.e., disabling loading the full dataset in RAM, and using the video files on the fly instead? Note that in order to use the ram_data functionality you need to run preprocessing scripts as in misc/pickle_frames.py.