Low-light denoising + 4x super-resolution (NAFNet)

Joint denoiser / 4x upscaler for noisy low-light images (256x160 -> 1024x640).

metric value
test-like holdout PSNR 37.131 dB
official val PSNR 39.381 dB
params 2.30M

Degradation model recovered from the data: LR = round(clip(box4(HR) + N(0, sqrt(2.514*I + 2.037)))).

Load

from huggingface_hub import hf_hub_download
import torch
ckpt = torch.load(hf_hub_download('piushdasss/nppe-lowlight-sr-nafnet', 'model.pt'), map_location='cpu')
Downloads last month
7
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support