Research map / Video / Video segmentation and tracking
CPGA: Coding Priors-Guided Aggregation Network for Compressed Video Quality Enhancement
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.
Proposes CPGA, a compressed video quality enhancement network that aggregates temporal and spatial information guided by coding priors (motion vectors, predictive frames, residual frames), and releases the VCP dataset of 300 HEVC-compressed videos with extracted coding priors.
Problem
Compressed video quality enhancement (VQE) methods ignore coding priors embedded in the bitstream (motion vectors, predictive frames, residual frames), and existing VQE datasets lack these priors.
Approach
Use motion vectors and predictive frames to guide inter-frame temporal aggregation, and the current residual frame to guide multi-scale non-local spatial aggregation, plus a shift-channel-attention quality enhancement module; train on a new dataset that provides coding priors extracted from HEVC bitstreams.
Key results
- Under LDB QP=37, CPGA achieves average ΔPSNR 0.82 dB, versus 0.79 dB for STDR and 0.76 dB for RFDA (about +0.03 dB over STDR).
- Ablation on coding priors: MVs, predictive frames and residual frames add 0.03, 0.05 and 0.05 dB respectively; all three together give 0.13 dB over the baseline (0.69 dB to 0.82 dB).
- CPGA runs at 10.42 FPS at 832x480 versus 8.77 for STDR, with 1386K parameters.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add motion vectors to the coding-prior-free baseline (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.72 | Helped |
| Add predictive frames to the coding-prior-free baseline (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.73 | Helped |
| Add residual frames to the coding-prior-free baseline (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.72 | Helped |
| Add motion vectors and predictive frames (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.77 | Helped |
| Add predictive and residual frames (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.79 | Helped |
| Add all three coding priors: motion vectors, predictive frames, and residual frames (ablation) | CPGA · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.69 -> 0.82 | Helped |
| Use CPGA instead of STDR | Compressed video quality enhancement · JCT-VC, LDB, QP 37, average of 18 sequences | ∆PSNR (dB): 0.79 -> 0.82 | Helped |
| Use CPGA instead of STDR | Compressed video quality enhancement · JCT-VC, RA, QP 37 | ∆PSNR (dB): 0.55 -> 0.57 | Helped |
Limitations
- Relies on coding priors being available from the bitstream, which limits applicability to content without such side information.
- Gains over STDR are modest (about 0.02-0.05 dB).
Experimental setup
Datasets: VCP MFQE JCT-VC test sequences
Benchmarks: JCT-VC standard test sequences (18 sequences) MFQE dataset
Metrics: ΔPSNR ΔSSIM FPS parameters
Baselines: MFQE STDF-R3L RFDA CF-STIF STDR
Builds on: STDF RFDA STDR CA-SDF
Closest papers by meaning
- RivuletMLP: An MLP-based Architecture for Efficient Compressed Video Quality Enhancement (CVPR 2025 · similarity 0.83)
- Multi-Frame Deformable Look-Up Table for Compressed Video Quality Enhancement (AAAI 2025 · similarity 0.83)
- Plug-and-Play Versatile Compressed Video Enhancement (CVPR 2025 · similarity 0.83)
- Sparse Video-Gen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity (ICML 2025 · similarity 0.83)
- Augmented Deep Contexts for Spatially Embedded Video Coding (CVPR 2025 · similarity 0.83)
- PMQ-VE: Progressive Multi-Frame Quantization for Video Enhancement (NeurIPS 2025 · 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.