rkhyper

Add --fp32

Umm, it worked for me on all input images. Yeah, it much slower compared to exe file, maybe exe used igpu too, and the code is not optimized for CPU...

I think the error is occured for face enhance (GFPGAN) , not for this model. Try removing this argument

Do you run inference_realesrgan.py or inference_realesrgan_video.py for video?

You can try with inference_realesrgan_video.py

You can convert the model to MNN (Alibaba). It support on android too.

Same with termux cli : ./realsr-ncnn -i input.jpeg -o output.jpg -m models-Real-ESRGAN from curl https://huggingface.co/spaces/tumuyan/RealSR/raw/main/install_realsr_android.sh | bash

python3 inference_realesrgan.py -n RealESRGAN_x4plus -i input.png -o output.png -s 2 --fp32 i have same problem