Research map / Image generation / Image restoration
Bricker to BRACE: A Bracket Exposure RAW Dataset and Restoration Model for Flicker-Banding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Builds Bricker, a synthetic-plus-real bracketed RAW flicker-banding dataset, and BRACE, a multi-frame restoration model with frequency-aware banding prior and cross-exposure attention, plus a new SFC metric.
Problem
Flicker-banding removal from screen-captured images is ill-posed for single frames because banding is hard to separate from rich screen texture; no bracketed RAW datasets exist for this task and simulators do not reproduce exposure-dependent stripe morphologies.
Approach
Exploit exposure bracketing in RAW: a ray-tracing physical screen simulator and an automated multi-exposure capture tool produce paired data, and a model injects a Fourier-derived banding prior into cross-frame spatial attention (FiLM modulation) on top of TRMNet-style recurrent fusion, with a two-stage synthetic-then-real training.
Key results
- BRACE reaches 28.44 PSNR on the synthetic test set and 28.15 PSNR on the real test set, the best of the compared methods.
- On the real test set BRACE achieves SFC 0.1847, versus 0.4036 for the second-best method (FlickerFormer).
- On the real test set BRACE reports LPIPS 0.1925 and DISTS 0.1113, both best among compared methods.
Surprising findings
- HDRFlow, designed for video HDR reconstruction, performs very poorly on this task (about 6.6 PSNR on both test sets).
- The synthetic-only model already removes real flicker-banding without structural failure, which the authors take as evidence the simulator matches real degradations.
Limitations
- Residual banding or color errors on extremely complex compound-jagged patterns or large brightness ranges.
- Device-dependent color bias when applied to cameras with different ISP parameters without fine-tuning.
- Multi-frame pipeline still has non-negligible computational overhead for on-device deployment.
Experimental setup
Datasets: Bricker synthetic Bricker real
Benchmarks: Bricker synthetic test Bricker real test
Metrics: PSNR SSIM LPIPS DISTS MS-SWD SFC
Baselines: Burstormer HDRFlow TMRNet Flickerformer
Builds on: TRMNet FiLM FlickerFormer Burstormer HDRFlow
Closest papers by meaning
- RIFLE: Removal of Image Flicker-Banding via Latent Diffusion Enhancement (NeurIPS 2026 · similarity 0.93)
- BurstDeflicker: A Benchmark Dataset for Flicker Removal in Dynamic Scenes (NeurIPS 2025 · similarity 0.88)
- Prune and Repaint: Content-Aware Image Retargeting for any Ratio (NeurIPS 2024 · similarity 0.84)
- Polarized Color Screen Matting (CVPR 2025 · similarity 0.83)
- PolarMatte: Fully Computational Ground-Truth-Quality Alpha Matte Extraction for Images and Video using Polarized Screen Matting (CVPR 2024 · similarity 0.83)
- Low-Latency Space-Time Supersampling for Real-Time Rendering (AAAI 2024 · similarity 0.82)
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.