Research map / Detection and segmentation / CNN features
Optimizing Video Object Detection via a Scale-Time Lattice
Read the paperCheck 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 |
|---|---|---|---|
| Use the complete ST-Lattice with adaptive key-frame selection | Faster R-CNN ResNet-101 video detection · ImageNet VID val | mAP: 79.6 | Helped |
| Use a faster key-frame selection tradeoff | ST-Lattice · ImageNet VID val | speed (fps): 20 -> 62 fps | Helped |
| Use MHI-based temporal propagation instead of linear interpolation or RGB-difference regression (ablation) | ST-Lattice propagation · ImageNet VID val, intermediate frames | mAP | Helped |
| Separate temporal motion estimation from spatial offset regression (ablation) | PRU · ImageNet VID val, intermediate-frame refinement | mAP: 72.1 -> 73.9 | Helped |
| Jointly train the propagation and refinement regressors (ablation) | Two-stage PRU · ImageNet VID val, intermediate-frame refinement | mAP: 73.7 -> 73.9 | Helped |
| Use a two-step propagation-and-refinement unit instead of a single-step regression unit (ablation) | PRU · ImageNet VID val, intermediate-frame refinement | mAP: 71.0 -> 73.9 | Helped |
| Increase spatial refinement network size from small to large while keeping propagation medium (ablation) | ST-Lattice · ImageNet VID val, fixed key-frame interval 24 | mAP after refinement: 72.5 -> 73.7 | Helped |
| Increase temporal propagation network size from small to large while keeping refinement medium (ablation) | ST-Lattice · ImageNet VID val, fixed key-frame interval 24 | mAP after refinement: 72.7 -> 73.5 | Helped |
| Select key frames adaptively rather than uniformly (ablation) | ST-Lattice · ImageNet VID val | mAP | Helped |
| Select key frames adaptively rather than uniformly (ablation) | Naive interpolation · ImageNet VID val | mAP | Helped |
| Use adaptive rather than uniform key frames (ablation) | Key-frame detector inputs · ImageNet VID val, selected key frames | mAP: 74.1 vs 73.3 | Hurt |
Closest papers by meaning
- Dynamic Zoom-In Network for Fast Object Detection in Large Images (CVPR 2018 · similarity 0.88)
- Training-Time-Friendly Network for Real-Time Object Detection (AAAI 2020 · similarity 0.88)
- Rethinking Scale-Aware Temporal Encoding for Event-based Object Detection (NeurIPS 2025 · similarity 0.87)
- Adaptive Object Detection Using Adjacency and Zoom Prediction (CVPR 2016 · similarity 0.87)
- Mobile Video Object Detection With Temporally-Aware Feature Maps (CVPR 2018 · similarity 0.87)
- MSTDiff: Multiscale-Aware Transformer Diffusion Network for Video Object Detection (AAAI 2026 · similarity 0.87)
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.