gQIR: Generative Quanta Image Reconstruction

Project Page | ArXiv | GitHub

### What You Can Run - **Single Frame (Stage-2):** Reconstruct one frame from either a clean GT image (internally simulated to SPAD) or a real SPAD frame. - **Burst (Stage-3):** Reconstruct from a fixed **77-frame** window using either GT videos/cubes or real photon cubes. - **Pipelines:** Toggle between **Color** and **Monochrome** reconstruction in both tabs. ### Supported Inputs - **Single GT:** Standard image uploads. - **Single Real:** Real SPAD frame image uploads. - **Burst GT:** Public-friendly videos (`.mp4`, `.mov`, `.wmv`, `.avi`, `.mkv`, `.webm`) plus research cube formats (`.npy`, `.npz`, `.pt`, `.h5`) or image folders. - **Burst Real:** Photon cubes (`.npy`, `.npz`, `.pt`, `.h5`) or image folders. ### Quick Usage 1. Pick pipeline and input mode. 2. Load input and select burst start index (for Stage-3). 3. Set PPP for GT simulation paths, then run reconstruction and compare input vs output side-by-side.