Winner of CVPR2026 NTIRE Challenge on Image Shadow Removal: Semantic and Geometric Guidance for Shadow Removal via Cascaded Refinement
Lorenzo Beltrame, Jules Salzinger, Filip Svoboda, Jasmin Lampert, Phillipp Fanta-Jende + 2 more
TLDR
This paper presents a three-stage, semantic and geometric-guided pipeline for image shadow removal, winning the CVPR2026 NTIRE challenge.
Key contributions
- A three-stage progressive pipeline for shadow removal, built on OmniSR, using iterative direct refinement.
- Integrates DINOv2 semantic guidance and geometric cues (depth, normals) across all refinement stages.
- Introduces a contraction-constrained objective to stabilize multi-stage optimization and reduce errors.
- Achieved first place in the CVPR2026 NTIRE WSRD+ challenge with state-of-the-art performance.
Why it matters
This paper introduces a state-of-the-art shadow removal method that won the CVPR2026 NTIRE challenge. By combining semantic and geometric guidance with a novel multi-stage refinement and optimization strategy, it significantly advances the field. This work provides a robust and effective solution for a challenging computer vision problem.
Original Abstract
We present a three-stage progressive shadow-removal pipeline for the CVPR2026 NTIRE WSRD+ challenge. Built on OmniSR, our method treats deshadowing as iterative direct refinement, where later stages correct residual artefacts left by earlier predictions. The model combines RGB appearance with frozen DINOv2 semantic guidance and geometric cues from monocular depth and surface normals, reused across all stages. To stabilise multi-stage optimisation, we introduce a contraction-constrained objective that encourages non-increasing reconstruction error across the cascade. A staged training pipeline transfers from earlier WSRD pretraining to WSRD+ supervision and final WSRD+ 2026 adaptation with cosine-annealed checkpoint ensembling. On the official WSRD+ 2026 hidden test set, our final ensemble achieved 26.680 PSNR, 0.8740 SSIM, 0.0578 LPIPS, and 26.135 FID, ranked first overall, and won the NTIRE 2026 Image Shadow Removal Challenge. The strong performance of the proposed model is further validated on the ISTD+ and UAV-SC+ datasets.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.