AI Paper Atlas

Research map / Image generation / Image restoration

From Shadow Generation To Shadow Removal

Zhihao Liu, Hui Yin, Xinyi Wu, Zhenyao Wu, Yang Mi, Song Wang

CVPR 2021 · Image generation › Image restoration · approach: Reflection removal · 127 citations (OpenAlex)

Check your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace patch-based self-supervision with generated paired shadow/non-shadow regionsG2R-ShadowNet vs. Le and Samaras · ISTD test, allRMSE: 4.0 -> 3.9Improved
Add paired shadow-free training data and skip shadow generationG2R-ShadowNet · ISTD test, allSSIM: 0.944 -> 0.957Improved
Add refinement subnet (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 12.3 -> 8.9Improved
Provide the shadow mask to the refinement subnet (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 12.5 -> 8.9Improved
Train the three subnetworks jointly with connected gradient flow (ablation)G2R-ShadowNet · ISTD test, shadow regionPSNR: 33.43 -> 33.58Mixed
Connect generator and removal subnet gradients (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 9.0 -> 8.8Improved
Add adversarial generation loss L_GAN (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 42.1 -> 8.9Improved
Add identity loss L_iden (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 43.1 -> 8.9Improved
Add shadow-removal loss L_rem (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 9.8 -> 8.9Improved
Add whole-image loss L_full (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 12.3 -> 8.9Improved
Add area loss L_area (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 9.3 -> 8.9Improved
Set area-loss dilation kernel tau to 50 rather than 0, 5, 15, or 100 (ablation)G2R-ShadowNet · ISTD test, shadow regionRMSE: 9.5 -> 8.9; 9.5 -> 8.9; 9.1 -> 8.9; 9.4 -> 8.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy