why center crop for testing instead of the whole image?

[zimenglan-sysu-512]

hi @junfu1115
i have a question: in BaseDataset, the func _val_sync_transform use center-crop to crop the image. is it reasonable to use center-crop to test the model instead of the whole image?
thanks.

[junfu1115]

@zimenglan-sysu-512 you can ref zhanghang1989/PyTorch-Encoding#127 (comment) and we adopt test.py to obtain final result

[zimenglan-sysu-512]

hi @junfu1115
in danet/test.py, it seems that u don't set any params for base_size and crop_size, and still use _val_sync_transform to process the image.
can u show some details how do u process the image and mask when u test?

[zihao-lu]

@zhiweifang he uses the whole image without sync_transform when testing

[zimenglan-sysu-512]

hi @Mr-zihao
can u show me the code? i can't find them. thanks.

[zihao-lu]

@zimenglan-sysu-512 in test.py, the mode of get_dataset_model is testval with which the images only are transformed into tensor without crop、flip etc...

[zihao-lu]

@zimenglan-sysu-512 And if you get the same result as the paper,please tell me,thanks,there are some troubles with me.

[zimenglan-sysu-512]

@Mr-zihao
u are right, i ignore this detail. thanks