Bug in test procedure

[cnnAndBn]

hi,@PkuRainBow
I trained a segfix model and as you state in https://github.com/openseg-group/openseg.pytorch/blob/master/MODEL_ZOO.md#segfix-1, I want to refine my segmentation result.
So as you said, first get the offset prediction and then refine.
But for the offset prediction, it reports:
openseg.pytorch-master/lib/datasets/tools/collate.py", line 108, in collate
assert pad_height >= 0 and pad_width >= 0
it seems to be the input image size larger than the size in configuration, some minus number occurs
"test": {
"batch_size": 16,
"mode": "ss_test",
"out_dir": "/msravcshare/dataset/seg_result/cityscapes",
"data_transformer": {
"size_mode": "fix_size",
"input_size": [2048, 1024],
"align_method": "only_pad"
}
},

[hsfzxjy]

Are you using a custom dataset?

[cnnAndBn]

yes