Research map / Video / Video segmentation and tracking
Skip-Convolutions for Efficient Video Processing
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply learned Skip-Conv gates to EfficientDet-D3 | EfficientDet-D3 · UA-DETRAC test | GMAC: 22.06 to 6.36 GMAC | Helped |
| Apply learned Skip-Conv gates to EfficientDet-D3 | EfficientDet-D3 · UA-DETRAC test | AP: 62.3 to 62.6 | Helped |
| Apply Skip-Conv to HRNet-w32 | HRNet-w32 · JHMDB | Avg PCK: 94.5 to 95.1 | Helped |
| Apply Skip-Conv to HRNet-w32 | HRNet-w32 · JHMDB | GMAC: 10.19 to 5.30 | Helped |
| Use learned Gumbel gates instead of norm-based gates (ablation) | Skip-Conv on EfficientDet · UA-DETRAC | GMAC | Helped |
| Use Output-Norm gates instead of Input-Norm gates (ablation) | Skip-Conv on EfficientDet · UA-DETRAC | GMAC | Helped |
| Use Output-Norm gates (ablation) | EfficientDet-D2 · UA-DETRAC | GMAC: 9.52 to 6.03 GMAC | Helped |
| Use structured gate blocks instead of unstructured 1×1 gates (ablation) | Skip-Conv on HRNet-w32 · JHMDB split 1 | PCK: 95.0 (1×1) and 94.9 (4×4); 95.0 (8×8) | No clear effect |
| Use 2×2 structured gate blocks (ablation) | Skip-Conv on HRNet-w32 · JHMDB split 1 | GMAC: 5.06 to 2.43 | Helped |
| Train and test with clip length 8 rather than train and test with clip length 4 (ablation) | Skip-Conv on HRNet-w32 · JHMDB split 1 | GMAC: 3.10 to 2.43 GMAC; 95.3 to 94.5 PCK | Mixed |
| Train with clip length 8 and test without periodic reference reset (ablation) | Skip-Conv on HRNet-w32 · JHMDB split 1 | PCK: 94.5 to 94.2 | Hurt |
| Apply Skip-Conv to TSN-25 | TSN-25 · Human action dataset split 1 RGB | Top-1 accuracy: 55.49 to 54.77 | Hurt |
Closest papers by meaning
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.87)
- The Surprising Effectiveness of Skip-Tuning in Diffusion Sampling (ICML 2024 · similarity 0.86)
- SkipSR: Faster Super-Resolution with Token Skipping (NeurIPS 2026 · similarity 0.86)
- Beyond Skip Connection: Pooling and Unpooling Design for Elimination Singularities (AAAI 2025 · similarity 0.85)
- AsymRnR: Video Diffusion Transformers Acceleration with Asymmetric Reduction and Restoration (ICML 2025 · similarity 0.85)
- Skip-It? Theoretical Conditions for Layer Skipping in Vision–Language Models (ICML 2026 · 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.