Toward Robust and 3D-Aware RGB-NIR Imaging in the Dark

Researchers proposed a 3D-aware RGB-NIR enhancement model that restores low-light images without clean RGB supervision.

The method fuses noisy RGB inputs with near-infrared cues in 3D space to improve reconstruction quality.

It shows strong results on both synthetic and real-world tests, and the code has been released.

The approach could make low-light imaging more robust while reducing the need for costly paired clean training data.