Low-Light NeRF Pipeline in PyTorch

I need an end-to-end neural rendering pipeline that can faithfully reconstruct low-light scenes with NeRF. All training will be image-based and must run entirely in PyTorch. You are free to choose or combine publicly available night-time or low-illumination datasets, as long as the licence allows research use and you clearly cite the source. Here is what I expect at delivery: • A reproducible PyTorch implementation (data loader, network, training loop, rendering code). • Training and inference scripts that run on a single GPU and are easy to adapt to new datasets. • Built-in evaluation that logs PSNR, SSIM and LPIPS after every validation epoch and prints a final summary table. • A short report (markdown or PDF) discussing dataset choice, hyper-parameters and final scores, plus a few rendered images demonstrating quality under low light. • Clean, commented code pushed to a Git repository. Acceptance will be based on the code running without modification and achieving competitive scores (≥ the median reported in the datasets you select) for all three metrics on at least one public low-light set.

Project ID: 40185069