Lingchen Sun

Thanks for your question. What is the resolution of your input? The SD-based SR methods are quite sensitive to the input resolution, which should ideally be larger than 512×512.

感谢你对我们的工作感兴趣！ 1. 是的，在SR任务中已有LR input 为强 image condition。在第二阶段的起始点加入 LR 同时也可以保持训测一致； 2. 在单卡的时候需要将args.is_module改为False。

Thanks for your feedback. Is the input image size 128x256x**3**? You may be able to resolve the issue by removing the 'tile_vae ' and 'tile_diffusion' when the input size is...

You can try: ''' python test_ccsr_tile.py ^ --image_path X:\input-CCSR2 ^ --pretrained_model_path ....\SD21-base\stable-diffusion-2-1-base ^ --controlnet_model_path ....\CCSR-CCSR-v2.0\preset\models\controlnet ^ --vae_model_path ....\CCSR-CCSR-v2.0\preset\models\vae ^ --sample_method ddpm --num_inference_steps 15 --t_min 0.0 --start_point lr --start_steps 0 --guidance_scale...

Thank you for your feedback! Currently, CCSR does not support other sampling methods. In CCSR-v2, we prioritized enhancing generative capacity compared to CCSR-v1. We plan to refine this balance by...

Thank you for your feedback! Currently, CCSR does not support other SD models. However, SD 2.1 can be directly replaced with other SD models within the Diffusers library. Since part...

The CCSR-v2 code and model have been updated and are now implemented using the Diffusers library. We look forward to hearing your feedback!

Thank you for your report. Could you first try commenting out or removing the import of mac_specific.py to see if it affects the execution? If the issue persists, you can...

Hello, I'm not entirely familiar with the detailed function of queue_size, but I believe it refers to the number of samples that can be preloaded into the training pool during...