Lingchen Sun

Hello, can you provide more information about your input LR image and the error message?

Thank you for your report. You can try adding the [mac_specific.py](https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/modules/mac_specific.py) from [Stable Diffusion WebUI](https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/master) to the corresponding code location.

From the error message, it seems that CUDA is not being used. Could you provide your CUDA version?

Thank you for your attention. The key difference lies in the weights applied when merging tile patches into the overall latent state. In inference_ccsr.py, average weights are used, whereas in...

Thank you for your attention! Yes, we utilize a combination of L1 loss, perceptual loss (LPIPS), and GAN loss to train the decoder. During training, we observed that optimizing the...

Compared to SeeSR, CCSR does not rely on extracting image semantic prompts to activate the generative capabilities of SD. However, the performance of SeeSR can be negatively affected by inaccurate...

Currently, we do not support multi-GPU inference. However, you can try using the tile mode([inference_ccsr_tile.py](https://github.com/csslc/CCSR/blob/CCSR-v1.0/inference_ccsr_tile.py) in CCSR-v1 and [test_ccsr_tile.py](https://github.com/csslc/CCSR/blob/CCSR-v2.0/test_ccsr_tile.py) in CCSR-v2 to reduce GPU memory usage during inference.

Hello, we have not used a prompt, as the current prompt extractor is large and sometimes inaccurate. However, we believe that developing a high-quality prompt extractor holds great potential and...

Thanks for your question. The weights of GAN loss `lambda_disc` and LPIPS loss `lambda_lpips` in stage 2 can be reduced for a better PSNR value.

Currently, batch inference is not supported. We plan to add this functionality next week.