TLC icon indicating copy to clipboard operation
TLC copied to clipboard

Published 20 hours ago • verified publisher icon megvii-research

thanks for your enjoy！i have a try in sidd dataset of HINet,but it gives me " TypeError: 'NoneType' object is not subscriptable",look for your reply,thank you!

Open zxcvads opened this issue 3 years ago • 4 comments

zxcvads avatar Dec 28 '21 02:12 zxcvads

Thank you for your feedback. We fixed the issue in the latest update. It is worth mentioning that, the training size and test size for HINet on SIDD dataset are both 256x256, so there is no train-test inconsistency as mentioned in the paper.

achusky avatar Dec 28 '21 04:12 achusky

Thank you for your feedback. We fixed the issue in the latest update. It is worth mentioning that, the training size and test size for HINet on SIDD dataset are both 256x256, so there is no train-test inconsistency as mentioned in the paper.

thanks for your reply! After your fixed code, i faced another bug: Traceback (most recent call last): File "basicsr/test_tlsc.py", line 68, in main() File "basicsr/test_tlsc.py", line 64, in main rgb2bgr=rgb2bgr, use_image=use_image) File "/code/HINet-main/basicsr/models/base_model.py", line 58, in validation save_img, rgb2bgr, use_image) File "/code/HINet-main/basicsr/models/image_restoration_model.py", line 323, in nondist_validation self.grids() #apply for demo.py File "/code/HINet-main/basicsr/models/image_restoration_model.py", line 132, in grids num_row = (h - 1) // crop_size + 1 TypeError: unsupported operand type(s) for //: 'int' and 'NoneType' 0%| | 0/1280 [00:00<?, ?image/s]

zxcvads avatar Dec 28 '21 06:12 zxcvads

Thank you for your feedback. We fixed the issue in the latest update. It is worth mentioning that, the training size and test size for HINet on SIDD dataset are both 256x256, so there is no train-test inconsistency as mentioned in the paper.

Thank you for your feedback. We fixed the issue in the latest update. It is worth mentioning that, the training size and test size for HINet on SIDD dataset are both 256x256, so there is no train-test inconsistency as mentioned in the paper.

Besides, my sidd datasets preparation is referenced by HINet, https://github.com/megvii-model/HINet/blob/main/scripts/data_preparation/sidd.py,

zxcvads avatar Dec 28 '21 07:12 zxcvads

Hi, perhaps you are using an older version of HINet code. You may set grids: false in config or update the code of image_restoration_model.py.

achusky avatar Dec 28 '21 08:12 achusky