Downloads · 30 days
0
dlp-26t2-nppe3-denoise-sr
thehimanshhu/dlp-26t2-nppe3-denoise-sr
dlp-26t2-nppe3-denoise-sr is a machine learning model from thehimanshhu. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
EDSR-style residual CNN (96 feats, 20 blocks, resscale 0.1) with two PixelShuffle(2) stages and a bilinear global skip. Input: noisy low-light 256x160 RGB. Output: clean 1024x640 RGB.
Access
Public
Updated Sep 6, 2026
Repo size
16.3 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt16.3 MB · 100%
At a glance
- License
- mit
- Access
- Public
- Created
- Sep 6, 2026
- Updated
- Sep 6, 2026
- SHA
- 71cf51e0
Model card
From the Hugging Face model README
DLP 26T2 NPPE3 — denoising + 4x super-resolution
EDSR-style residual CNN (96 feats, 20 blocks, res_scale 0.1) with two PixelShuffle(2) stages and a bilinear global skip. Input: noisy low-light 256x160 RGB. Output: clean 1024x640 RGB.
Trained with Charbonnier loss, AdamW (0.0002 -> 1e-06, cosine), AMP, 64px LR crops, batch 32, on 2x T4.
Best validation PSNR (RGB): 38.711 dB
import torch
ck = torch.load('best.pt', map_location='cpu')
model = DenoiseSRNet(96, 20, 0.1, 4)
model.load_state_dict(ck['state_dict'])
- License
- mit
- Created
- Sep 6, 2026
- Updated
- Sep 6, 2026