Research map / Detection and segmentation / CNN features
Sequence Level Semantics Aggregation for Video Object Detection
Haiping Wu, Yuntao Chen, Naiyan Wang, Zhaoxiang Zhang
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 semantics aggregation within a single frame (ablation) | single-frame Faster R-CNN · ImageNet VID val | mAP: 73.62 -> 75.26 | Improved |
| add semantics aggregation within a single frame (ablation) | single-frame Faster R-CNN · ImageNet VID val (fast motion) | mAP: 51.53 -> 51.43 | No clear effect |
| extend semantic neighbor sampling to full video sequence (ablation) | single-frame semantics aggregation · ImageNet VID val | mAP: 75.26 -> 80.25 | Improved |
| extend semantic neighbor sampling to full video sequence (ablation) | single-frame semantics aggregation · ImageNet VID val (fast motion) | mAP: 51.43 -> 61.38 | Improved |
| use 21 frames instead of 5 for aggregation (ablation) | SELSA · ImageNet VID val | mAP: +1.04 mAP | Improved |
| increase sampling stride from 1 to 10 (ablation) | SELSA 21 frames · ImageNet VID val | mAP: 77.02 -> 79.36 | Improved |
| shuffled sampling of semantic neighbors instead of stride 10 (ablation) | SELSA 21 frames · ImageNet VID val | mAP: 79.36 -> 80.25 | Improved |
| add data augmentation (ablation) | SELSA ResNet-101 · ImageNet VID val | mAP: +2.44 mAP | Improved |
| add data augmentation (ablation) | SELSA ResNeXt-101 · ImageNet VID val | mAP: 83.11 -> 84.30 | Improved |
| add Seq-NMS post-processing (ablation) | SELSA ResNet-101 · ImageNet VID val | mAP: 82.69 -> 82.48 | Hurt |
| add Seq-NMS post-processing (ablation) | SELSA ResNeXt-101 · ImageNet VID val | mAP: 84.30 -> 83.73 | Hurt |
| use ResNeXt-101 instead of ResNet-101 backbone | SELSA · ImageNet VID val | mAP: 80.25 -> 83.11 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Identity-Consistent Aggregation for Video Object Detection (ICCV 2023 · similarity 0.88)
- Practical Video Object Detection via Feature Selection and Aggregation (IJCV 2026 · similarity 0.87)
- Feature Aggregated Queries for Transformer-Based Video Object Detectors (CVPR 2023 · similarity 0.87)
- YOLOV: Making Still Image Object Detectors Great at Video Object Detection (AAAI 2023 · similarity 0.86)
- Video Object Detection via Object-level Temporal Aggregation (ECCV 2020 · similarity 0.86)
- Not All Tokens Matter All The Time: Dynamic Token Aggregation Towards Efficient Detection Transformers (ICML 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.
