Research map / Video / Video segmentation and tracking
Non-Semantics Suppressed Mask Learning for Unsupervised Video Semantic Compression
Yuan Tian, Guo Lu, Guangtao Zhai, Zhiyong Gao
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 |
|---|---|---|---|
| add semantic compensation (residual semantic stream) to VVC (ablation) | SMC · UCF101 TSM @0.02bpp | Top1 accuracy (%): 16% Top1 drop without compensation | Improved |
| use video enhancement BasicVSR++ instead of semantic compensation (ablation) | SMC · UCF101 TSM @0.02bpp | Top1 accuracy (%): -14.23%@0.02bpp | Hurt |
| add NSS entropy suppression loss to MAE objective (ablation) | SMC · UCF101 TSM (VVC CRF 51) | Top1 accuracy (%): 69.57 -> 72.96 | Improved |
| add NSS entropy suppression loss to MAE objective (ablation) | SMC · UCF101 TSM (VVC CRF 51) | residual bitrate (bpp): 0.0074 -> 0.0028 | Improved |
| condition GMM on mined semantic feature S instead of a plain learnable variable (ablation) | NSS · UCF101 TSM (VVC CRF 51) | Top1 accuracy (%): 71.38 -> 72.96 | Improved |
| use GMM entropy regularization instead of VQ codebook for token discretization (ablation) | NSS · UCF101 TSM (VVC CRF 51) | Top1 accuracy (%): 71.22 vs 71.38 (VQ vs woSem) | No clear effect |
| use NSS-MAE objective instead of MoCoV3 contrastive objective (ablation) | SMC · MOT17 ByteTrack @0.01bpp | MOTA (%): 67.62 -> 70.84 | Improved |
| use NSS-MAE objective instead of MoCoV3 (ablation) | SMC · UCF101 TSM @0.02bpp | Top1 accuracy (%): 75.20 -> 76.48 | Improved |
| fine-tune downstream TSM on videos from each codec | VVC · UCF101 TSM @0.04bpp | Top1 accuracy (%): 76.97 -> 82.35 | Improved |
| fine-tune downstream TSM on SMC-decoded videos | SMC · UCF101 TSM @0.04bpp | Top1 accuracy (%): 86.46 -> 89.83 | Improved |
| replace NSS-MAE semantic objective with hand-crafted SegMap loss (ablation) | SMC · UCF101 TSM | Top1 accuracy (%): about 3% gap at 0.05bpp | Hurt |
| set NSS weight beta to 0.01 or 1 instead of 0.1 (ablation) | SMC · UCF101 TSM | Top1 accuracy (%) | Hurt |
Closest papers by meaning
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.87)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.86)
- Towards Unified Human Perception and Machine Understanding: Token Flow Guided Compression Framework (CVPR 2026 · similarity 0.86)
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.85)
- PVC: Progressive Visual Token Compression for Unified Image and Video Processing in Large Vision-Language Models (CVPR 2025 · similarity 0.85)
- Ultra Lowrate Image Compression with Semantic Residual Coding and Compression-aware Diffusion (ICML 2025 · similarity 0.85)
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.
