low-light-super-resolution

Low-Light Denoising, Deblurring & 4x Super-Resolution

This repository contains the trained weights and architecture for MPR_SRNet (Multi-Stage Progressive Restoration and RCAB Super-Resolution Network).

Architecture Details

• Stage 1 (Restoration): 5 Non-linear Activation Free (NAF) / Gated blocks for low-light enhancement, denoising, and deblurring.
• Stage 2 (Super-Resolution): 8 Residual Channel Attention Blocks (RCAB) for feature extraction.
• Upscaling Head: 4x PixelShuffle upsampler.

How to Load the Model in PyTorch

import torch
from model import MPR_SRNet

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 1. Initialize model architecture
model = MPR_SRNet(
    in_channels=3, 
    out_channels=3, 
    feature_dim=64, 
    num_naf_blocks=5, 
    num_rcab_blocks=8, 
    scale_factor=4
)

# 2. Load trained weights
checkpoint = torch.load("best_model.pth", map_location=device)
state_dict = checkpoint.get("model_state_dict", checkpoint)
model.load_state_dict(state_dict)

model.to(device)
model.eval()
print("Model loaded successfully!")