caiyuanhao1998/Retinexformer
Transformer-based model for low-light image enhancement that won NTIRE 2025 and 2026 challenges.

Velocity · 7d
+1.4
★ / day
Trend
→steady
star history
Retinexformer is a one-stage transformer-based approach for low-light image enhancement based on Retinex theory. The model won runner-up in NTIRE 2024, first place in NTIRE 2025, and first place in NTIRE 2026 Efficient Low-light Image Enhancement Challenge. The repository provides a baseline and toolbox supporting over 15 benchmarks and handles extremely high-resolution images up to 4000x6000 pixels.