Changing the batch_size in test_tryon doesn't work

[tobias861222]

Thank you for the wonderful work. I tried changing the batch_size in test_tryon.py during the inference, but it didn't have any effect for my result. I use only one GPU(1080ti) and run on Ubuntu20.04 LTS , so i made a few change for the code.

1, delete distributed function torch.distributed.init_process_group( 'nccl', init_method='env://' )

2,

train_sampler = DistributedSampler(train_data)
train_loader = DataLoader(train_data, batch_size=opt.batchSize, shuffle=False,
                                               num_workers=4, pin_memory=True, sampler=train_sampler)

change to

train_loader = DataLoader(train_data, batch_size=opt.batchSize, shuffle=False,
                                               num_workers=4, pin_memory=True) 

3,

gen_model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(gen_model).to(device)
if opt.isTrain and len(opt.gpu_ids):
    model_gen = torch.nn.parallel.DistributedDataParallel(gen_model, device_ids=[opt.local_rank])

change to model_gen = gen_model.to(device)

Here is my .sh file The result remains the same regardless of whether I change the batch size to 1 or 64.

python3 test_tryon_v2.py \
        --name test_gpvtongen_vitonhd_unpaired_1109 \
        --resize_or_crop None --verbose --tf_log \
        --batchSize 64 \
        --num_gpus 1 --label_nc 14 --launcher pytorch \
        --PBAFN_gen_checkpoint 'checkpoints/gp-vton_gen_vitonhd_wskin_wgan_lrarms_1029/PBAFN_gen_epoch_201.pth' \
        --dataroot ./dataset\
        --image_pairs_txt test_pairs_unpaired_1018.txt \
        --warproot ./sample/test_partflow_vitonhd_unpaired_1109        

However, when I change gen_model.train() to gen_model.eval(), the result remains the same regardless of how I modify the batch size.

"I think the problem might be that the batch size wasn't properly changed. Can someone help me resolve this issue? Thanks~