Research map / Video / Video segmentation and tracking
Non-Local ConvLSTM for Video Compression Artifact Reduction
Yi Xu, Longwen Gao, Kai Tian, Shuigeng Zhou, Huyang Sun
Read the paperCheck 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 changed | Where | Measured | Effect |
|---|---|---|---|
| use 7 frames with ConvLSTM instead of a single frame (ablation) | encoder-decoder · Yang et al. QP=37 | dPSNR: 0.395 -> 0.456 | Improved |
| replace plain ConvLSTM with non-local ConvLSTM (ablation) | ConvLSTM with 7 frames · Yang et al. QP=37 | dPSNR: 0.456 -> 0.601 | Improved |
| use non-local module instead of motion estimation and compensation (ablation) | ME-ConvLSTM · Yang et al. QP=37 | dPSNR: 0.503 -> 0.601 | Improved |
| use non-local module instead of motion estimation and compensation (ablation) | ME-ConvLSTM · Yang et al. QP=37 | dSSIM: 0.827 -> 0.897 | Improved |
| fine-tune on longer 20-frame sequences (ablation) | NL-ConvLSTM · Yang et al. QP=37 | dPSNR: 0.601 -> 0.604 | Improved |
| two-stage approximation of non-local similarity vs original non-local (ablation) | NL-ConvLSTM · run-time at 180x180 | ms per frame: 4391.75 -> 102.13 | Improved |
| use multiple preceding and following frames with non-local ConvLSTM instead of two PQFs (MFQE) | NL-ConvLSTM · Yang et al. QP=37 | dPSNR: 0.510 -> 0.601 | Improved |
| use NL-ConvLSTM instead of MFQE | NL-ConvLSTM · Yang et al. QP=42 | dPSNR: 0.461 -> 0.614 | Improved |
| NL-ConvLSTM instead of 3D CNN | NL-ConvLSTM · Vimeo-90K QP=37 | PSNR: 33.25 -> 33.39 | Improved |
| NL-ConvLSTM instead of MFQE on quality fluctuation | NL-ConvLSTM · Yang et al. sequences | PSNR STD: 1.055 -> 1.036 | Improved |
| NL-ConvLSTM instead of MFQE on peak-valley difference | NL-ConvLSTM · Yang et al. sequences | PSNR PVD: 1.109 -> 1.038 | Improved |
| use NL-ConvLSTM, which uses more frames, instead of single-frame or pair methods | NL-ConvLSTM · run-time 1920x1080 | ms per frame: 1009.00 -> 6738.00 | Hurt |
Closest papers by meaning
- Neural Video Compression with Feature Modulation (CVPR 2024 · similarity 0.85)
- ECVC: Exploiting Non-Local Correlations in Multiple Frames for Contextual Video Compression (CVPR 2025 · similarity 0.84)
- NVRC: Neural Video Representation Compression (NeurIPS 2024 · similarity 0.84)
- Combining Frame and GOP Embeddings for Neural Video Representation (CVPR 2024 · similarity 0.84)
- Towards Practical Real-Time Neural Video Compression (CVPR 2025 · similarity 0.84)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · 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.
