| tags: | |
| - image-super-resolution | |
| - image-denoising | |
| - pytorch | |
| # Low-Light Denoising + 4x Super-Resolution | |
| Compact RRDB-style CNN trained for the DLP 26T2 NPPE3 Kaggle competition. Takes a noisy, | |
| low-light, low-resolution RGB image and outputs a denoised 4x super-resolved image. | |
| - Params: 6.13M | |
| - Best validation PSNR: 38.86 dB | |
| - Architecture: 8 RRDB blocks, 64 features, | |
| pixel-shuffle upsampling, trained with L1 loss. | |
| ## Usage | |
| ```python | |
| import torch | |
| from model import DenoiseSRNet # architecture defined in this notebook | |
| model = DenoiseSRNet(nf=64, gc=32, | |
| n_blocks=8, scale=4) | |
| model.load_state_dict(torch.load("best.pth", map_location="cpu")) | |
| model.eval() | |
| ``` | |