patel420/dlp-26t2-nppe3-denoise-sr

DenoiseSRNet (RRDB) — Joint Denoising + 4x Super-Resolution

Trained for the DLP 26T2 NPPE3 Kaggle competition.

• Architecture: RRDB backbone (12 blocks, 64 channels, growth=32)
• Best validation PSNR: 39.089 dB

Usage

import torch, json
from huggingface_hub import hf_hub_download
import importlib.util

config_path = hf_hub_download(repo_id="patel420/dlp-26t2-nppe3-denoise-sr", filename="config.json")
weights_path = hf_hub_download(repo_id="patel420/dlp-26t2-nppe3-denoise-sr", filename="pytorch_model.bin")
modeling_path = hf_hub_download(repo_id="patel420/dlp-26t2-nppe3-denoise-sr", filename="modeling.py")

spec = importlib.util.spec_from_file_location("modeling", modeling_path)
modeling = importlib.util.module_from_spec(spec)
spec.loader.exec_module(modeling)

config = json.load(open(config_path))
model = modeling.DenoiseSRNet(**config)
model.load_state_dict(torch.load(weights_path, map_location="cpu"))
model.eval()