Research map / Video / Video segmentation and tracking
High Dynamic Range Video Compression: A Large-Scale Benchmark Dataset and A Learned Bit-depth Scalable Compression Algorithm
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.
Collects HDRVD2K, a 2200-clip HDR video training dataset, and proposes LBSVC, a learned bit-depth scalable HDR video codec built on DCVC-HEM with a dynamic-range-prior bit-depth enhancement module.
Problem
Learned video compression for HDR video is unexplored because no large, high-quality HDR video training data exists, and existing LVC methods ignore the dynamic range gap between HDR and tone-mapped LDR videos.
Approach
A large HDR dataset (HDRVD2K) plus a two-layer bit-depth scalable LVC where a bit-depth enhancement module predicts 16-bit HDR features from compressed 8-bit tone-mapped LDR features guided by a lightweight histogram-based dynamic range prior.
Key results
- LBSVC achieves 32.5% bitrate saving over SHM-12.4 in PU-SSIM on the HDM dataset (Figure 1).
- Table 3 BD-Rate: LBSVC reaches -45.2% (JVET) and -49.8% (HDM) on PU-PSNR relative to SHM.
- Removing the BEM raises BD-Rate by about 8.5/7.5 (PU-PSNR) on JVET/HDM, showing its contribution.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use LBSVC instead of SHM | HDR video compression · JVET | BD-Rate (PU-PSNR): -45.2 | Helped |
| Use LBSVC instead of SHM | HDR video compression · HDM | BD-Rate (PU-PSNR): -49.8 | Helped |
| Use LBSVC instead of SHM | HDR video compression · JVET | BD-Rate (PU-SSIM): -93.9 | Helped |
| Use LBSVC instead of SHM | HDR video compression · HDM | BD-Rate (PU-SSIM): -32.5 | Helped |
| Use LBSVC instead of SHM | HDR video compression · JVET | BD-Rate (VDP2): -9.0 | Helped |
| Use LBSVC instead of SHM | HDR video compression · HDM | BD-Rate (VDP2): -7.9 | Helped |
| Use LBSVC instead of SHM | HDR video compression · JVET | BD-Rate (VDP3): -15.5 | Helped |
| Use LBSVC instead of SHM | HDR video compression · HDM | BD-Rate (VDP3): -44.2 | Helped |
| Add BEM (ablation) | LBSVC · JVET | BD-Rate (PU-PSNR): 8.51 | Helped |
| Add BEM (ablation) | LBSVC · HDM | BD-Rate (PU-PSNR): 7.49 | Helped |
| Add BEM (ablation) | LBSVC · JVET | BD-Rate (PU-SSIM): 8.05 | Helped |
| Add BEM (ablation) | LBSVC · HDM | BD-Rate (PU-SSIM): 7.11 | Helped |
Surprising findings
- LVC methods lose their advantage on HDR-VDP-2 and HDR-VDP-3, where HEM* and LSSVC have positive BD-Rate versus SHM.
Limitations
- Evaluation limited to 16 test sequences from JVET and HDM; BL+EL comparisons only against a few baselines.
- VDP metrics show LVC methods underperform SHM in several conditions.
Experimental setup
Datasets: HDRVD2K JVET HDM
Benchmarks: JVET HDR test sequences HDM
Metrics: PU-PSNR PU-SSIM HDR-VDP-2 HDR-VDP-3 BD-Rate bpp
Baselines: SHM-12.4 Mai11 HEM* LSSVC
Builds on: DCVC-HEM LSSVC SHM BD-Rate
Closest papers by meaning
- DeAltHDR: Learning HDR Video Reconstruction from Degraded Alternating Exposure Sequences (ICLR 2026 · similarity 0.88)
- Towards Real-World HDR Video Reconstruction: A Large-Scale Benchmark Dataset and A Two-Stage Alignment Network (CVPR 2024 · similarity 0.87)
- HDRFlow: Real-Time HDR Video Reconstruction with Large Motions (CVPR 2024 · similarity 0.87)
- EvHDR-GS: Event-guided HDR Video Reconstruction with 3D Gaussian Splatting (AAAI 2025 · similarity 0.86)
- FLAVC: Learned Video Compression with Feature Level Attention (CVPR 2025 · similarity 0.84)
- Neural Video Compression with Feature Modulation (CVPR 2024 · similarity 0.84)
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.