search

caiyuanhao1998 / Retinexformer

"Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement" (ICCV 2023) & (NTIRE 2024 Challenge)
https://arxiv.org/abs/2303.06705
MIT License
897 stars 78 forks source link

Ask about dataset about sid #39

Closed YanJieWen closed 9 months ago

YanJieWen commented 9 months ago

I would like to ask about the data set SID, gtsize is defined as 256, but the read Dataset is [512,960] and gt_sizes is 256. Is the input gt_sizes defined in this way invalid? The size of the input network is 512,960. This is for a single block. Is the GPU taxing too much? Or the picture has been scaled somewhere in the code, and I hope the author will point it out. Thanks~

caiyuanhao1998 commented 9 months ago

we crop patches of 256x256 during training to save the computational costs. In the testing phase, we use the whole image with its original size.

If you find our repo helpful, please help us star it.

Thanks.