search

Fayeben / GenerativeDiffusionPrior

Generative Diffusion Prior for Unified Image Restoration and Enhancement (CVPR2023)
Apache License 2.0
277 stars 30 forks source link

colorization task not work well #6

Closed chensming closed 1 year ago

chensming commented 1 year ago

Hi, bro, nice jobs! I am interested in it and I try I repeat the experiment. When I run these script: MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True" python sample_x0_color.py $MODEL_FLAGS --use_img_for_guidance --start_from_scratch --save_png_files It doesn't work in some pictures. could you help me about it? In the pic, the first example seems work, but the orders didn't. (The pictures come from imagenet/val/n03781244) 1683616000784

Fayeben commented 1 year ago

Thanks! Since colorization is not a trivial task, maybe you can try to modify the weight of loss and guidance scale to solve this problem. https://github.com/Fayeben/GenerativeDiffusionPrior/blob/116b875db9b4b6c3bcc8adcfc7dbaedb607e1edb/scripts/sample_x0_color.py#L112 https://github.com/Fayeben/GenerativeDiffusionPrior/blob/116b875db9b4b6c3bcc8adcfc7dbaedb607e1edb/scripts/sample_x0_color.py#L252

chensming commented 1 year ago

ok. I will give it a try, thanks!

Fayeben commented 1 year ago

Since there are no more questions, I'll mark it closed. If you still have some questions, please feel free to ask. Thanks!