search

IDKiro / DehazeFormer

[IEEE TIP] Vision Transformers for Single Image Dehazing
MIT License
369 stars 35 forks source link

test datasets #8

Closed ShenZhen0502 closed 2 years ago

ShenZhen0502 commented 2 years ago

Good job! Does the experiment in your paper need to resize the test image to the same size as the training image?

IDKiro commented 2 years ago

No, you can feed images of any resolution during testing. Also, we don't resize the images during training, just crop them to 256x256 patches.

IDKiro commented 2 years ago

Reopen it if necessary.

Algolzw commented 1 year ago

Hi, do you crop all images to 256x256 in both training and testing (as shown in the PairLoader)?

IDKiro commented 1 year ago

Hi, do you crop all images to 256x256 in both training and testing (as shown in the PairLoader)?

No, see lines 79~83 in loader.py. When setting self.mode = 'test', there is no cropping.