Research map / Detection and segmentation / CNN features
Practical Video Object Detection via Feature Selection and Aggregation
Yuheng Shi, Tong Zhang, XiaoJie Guo
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 feature selection and aggregation (TopK+NMS) to the base detector (ablation) | YOLOX-S · ImageNet VID val | AP50: 69.5 -> 75.4 | Improved |
| add affinity matrix to the aggregation (ablation) | YOLOV (TopK+NMS) · ImageNet VID val | AP50: 75.4 -> 76.9 | Improved |
| add average pooling over reference features (ablation) | YOLOV (TopK+NMS) · ImageNet VID val | AP50: 76.9 -> 77.3 | Improved |
| add regression-branch aggregation FAM_r on top of TopK+NMS (ablation) | YOLOV++ · ImageNet VID val | AP50: 77.3 -> 75.3 | Hurt |
| use Thresh instead of TopK+NMS together with FAM_r (ablation) | YOLOV++ · ImageNet VID val | AP50: 75.3 -> 78.7 | Improved |
| add FAM_r under Thresh selection (ablation) | YOLOV++ · ImageNet VID val | AP50: 76.4 -> 78.7 | Improved |
| use global instead of local reference frame sampling (4 global vs 39 local) (ablation) | YOLOV++-S · ImageNet VID val | AP50: 74.1 -> 75.7 | Improved |
| increase global reference frames from 4 to 31 (ablation) | YOLOV++-S · ImageNet VID val | AP50: 75.7 -> 78.7 | Improved |
| raise average-pooling threshold tau from 0 to 0.75 (ablation) | YOLOV++-S · ImageNet VID val | AP50: 77.1 -> 78.7 | Improved |
| set tau to 1 (no pooling selection) vs 0.75 (ablation) | YOLOV++-S · ImageNet VID val | AP50: 78.7 -> 78.3 | Hurt |
| upgrade YOLOV to YOLOV++ | YOLOX-S · ImageNet VID val | AP50: 77.3 -> 78.7 | Improved |
| apply the strategy to PPYOLOE-S | PPYOLOE-S · ImageNet VID val | AP50: 69.5 -> 75.6 | 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.
- PTSEFormer — PTSEFormer: Progressive Temporal-Spatial Enhanced TransFormer towards Video Object Detection (ECCV 2022) · better on 1 of 1 shared result
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
- YOLOV: Making Still Image Object Detectors Great at Video Object Detection (AAAI 2023 · similarity 0.92)
- Flow-Guided Feature Aggregation for Video Object Detection (ICCV 2017 · similarity 0.91)
- Video Object Detection via Object-level Temporal Aggregation (ECCV 2020 · similarity 0.90)
- Efficient One-Stage Video Object Detection by Exploiting Temporal Consistency (ECCV 2022 · similarity 0.90)
- Feature Aggregated Queries for Transformer-Based Video Object Detectors (CVPR 2023 · similarity 0.90)
- Context Enhanced Transformer for Single Image Object Detection in Video Data (AAAI 2024 · similarity 0.86)
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.
