multi-gpu training about flownet2-pytorch

[fitsumreda]

@LiliMeng could you give more details. what batch size was used, and which architecture did you try?

[ericzw]

@fitsumreda I trained flownet2 model with 1 GPU. there is one mistake like the follow picture. but it runs in epoch 0. it go wrong when validated after trained in 1 epoch.

in other hands, when I run this command to trained with 4 GPU, it can't work.
python main.py --start_epoch 0 --total_epochs 10 --batch_size 8 --model FlowNet2 --loss=L1Loss --optimizer=Adam --optimizer_lr=1e-10 --number_workers 0 --number_gpus 4
--training_dataset MpiSintelFinal --training_dataset_root /var/storage/shared/pnrsy/sys/jobs/application_1531419344053_9844/flownet2_data/training
--validation_dataset MpiSintelClean --validation_dataset_root /var/storage/shared/pnrsy/sys/jobs/application_1531419344053_9844/flownet2_data/training --resume /var/storage/shared/pnrsy/sys/jobs/application_1531419344053_9844/flownet2_data/pre-trained_model/FlowNet2_checkpoint.pth

could you help me to solve this problem?