Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Shadows: A Large-Scale Benchmark and Multi-Stage Framework for High-Fidelity Facial Shadow Removal

Tailong Luo, Jiesong Bai, Jinyang Huang, Junyu Xia, Wangyu Wu, Xuhang Chen

TL;DR

This work addresses the challenge of removing facial shadows while preserving texture under real-world lighting. It introduces ASFW, a large-scale real-world dataset of 1,081 aligned shadow/shadow-free pairs, and the Face Shadow Eraser (FSE), a three-stage framework consisting of mask-guided shadow localization, coarse shadow removal, and facial-aware refinement. The approach demonstrates improvements on real-world data and provides strong quantitative and qualitative results, with extensive ablations validating each component. By bridging synthetic and real domains, ASFW and FSE offer a practical pathway for robust facial shadow removal in downstream vision tasks.

Abstract

Facial shadows often degrade image quality and the performance of vision algorithms. Existing methods struggle to remove shadows while preserving texture, especially under complex lighting conditions, and they lack real-world paired datasets for training. We present the Augmented Shadow Face in the Wild (ASFW) dataset, the first large-scale real-world dataset for facial shadow removal, containing 1,081 paired shadow and shadow-free images created via a professional Photoshop workflow. ASFW offers photorealistic shadow variations and accurate ground truths, bridging the gap between synthetic and real domains. Deep models trained on ASFW demonstrate improved shadow removal in real-world conditions. We also introduce the Face Shadow Eraser (FSE) method to showcase the effectiveness of the dataset. Experiments demonstrate that ASFW enhances the performance of facial shadow removal models, setting new standards for this task.

Beyond Shadows: A Large-Scale Benchmark and Multi-Stage Framework for High-Fidelity Facial Shadow Removal

TL;DR

This work addresses the challenge of removing facial shadows while preserving texture under real-world lighting. It introduces ASFW, a large-scale real-world dataset of 1,081 aligned shadow/shadow-free pairs, and the Face Shadow Eraser (FSE), a three-stage framework consisting of mask-guided shadow localization, coarse shadow removal, and facial-aware refinement. The approach demonstrates improvements on real-world data and provides strong quantitative and qualitative results, with extensive ablations validating each component. By bridging synthetic and real domains, ASFW and FSE offer a practical pathway for robust facial shadow removal in downstream vision tasks.

Abstract

Facial shadows often degrade image quality and the performance of vision algorithms. Existing methods struggle to remove shadows while preserving texture, especially under complex lighting conditions, and they lack real-world paired datasets for training. We present the Augmented Shadow Face in the Wild (ASFW) dataset, the first large-scale real-world dataset for facial shadow removal, containing 1,081 paired shadow and shadow-free images created via a professional Photoshop workflow. ASFW offers photorealistic shadow variations and accurate ground truths, bridging the gap between synthetic and real domains. Deep models trained on ASFW demonstrate improved shadow removal in real-world conditions. We also introduce the Face Shadow Eraser (FSE) method to showcase the effectiveness of the dataset. Experiments demonstrate that ASFW enhances the performance of facial shadow removal models, setting new standards for this task.
Paper Structure (18 sections, 7 equations, 3 figures, 2 tables)

This paper contains 18 sections, 7 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. ASFW dataset
  3. Challenges in Dataset Construction
  4. Data Construction Pipeline
  5. Shadowed Image Creation Pipeline
  6. Shadow-Free Image Creation Pipeline
  7. Method
  8. Overall Framework
  9. MaskGuideNet: Shadow-Aware Mask Generation
  10. CoarseGenNet: Coarse Shadow Removal
  11. RefineFaceNet: Facial-Aware Refinement
  12. Experiment
  13. Implementation Details
  14. Comparisons with other Methods
  15. Qualitative Results
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Construction process of the ASFW dataset. The bidirectional workflow includes shadow synthesis (top) and shadow removal (bottom), leveraging Adobe Photoshop tools for photorealistic alignment.
  • Figure 2: Model overview.
  • Figure 3: Visual comparison of facial shadow removal on ASFW and UCB dataset.