Toward Robust and 3D-Aware RGB-NIR Imaging in the Dark
The paper claims clean RGB supervision is not needed to recover color images from extremely noisy RGB plus NIR input.
Its method uses 3D-aware neural modeling to fuse RGB observations with near-infrared cues in 3D space. The authors frame this as a way around carefully curated clean RGB training pairs, which current RGB-NIR low-light methods often require. They report stronger results on both synthetic and real data, with generalization across different noise levels. Codes and models are listed as available. HF Daily Papers' note
Its method uses 3D-aware neural modeling to fuse RGB observations with near-infrared cues in 3D space. The authors frame this as a way around carefully curated clean RGB training pairs, which current RGB-NIR low-light methods often require. They report stronger results on both synthetic and real data, with generalization across different noise levels. Codes and models are listed as available. HF Daily Papers' note
score 4