Research map / Detection and segmentation / CNN features
A Delay Metric for Video Object Detection: What Average Precision Fails to Tell
Huizi Mao, Xiaodong Yang, William J. Dally
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 flow-guided feature aggregation (FGFA) | R-FCN ResNet-101 · VIDT | mAP: 0.642 -> 0.675 | Improved |
| add flow-guided feature aggregation (FGFA) | R-FCN ResNet-101 · VIDT | average delay (AD): 9.0 -> 10.2 | Hurt |
| aggregate more frames in FGFA | FGFA · VIDT | mAP and AD | Mixed |
| use sparse key frames (DFF, 1 in 10) instead of per-frame R-FCN half model | R-FCN · VIDT | average delay (AD): 11.2 -> 11.6 | Hurt |
| add tracker-driven cascade (CaTDet) | Faster R-CNN ResNet-50 · VIDT | average delay (AD): 8.2 -> 9.2 | Hurt |
| add tracker-driven cascade (CaTDet) | Faster R-CNN ResNet-50 · VIDT | mAP: 0.561 -> 0.555 | No clear effect |
| increase anchor scales from 2 to 3 (ablation) | Faster R-CNN ResNet-50 · VIDT | average delay (AD), overall: 9.9 -> 8.8 | Improved |
| increase anchor scales from 3 to 4 (add small scale) (ablation) | Faster R-CNN ResNet-50 · VIDT | average delay (AD), overall: 8.8 -> 8.2 | Improved |
| increase anchor scales from 3 to 4 (ablation) | Faster R-CNN ResNet-50 · VIDT | mAP: 0.563 -> 0.562 | No clear effect |
| increase anchor scales from 4 to 5 (ablation) | Faster R-CNN ResNet-50 · VIDT | average delay (AD), overall: 8.2 -> 8.4 | No clear effect |
| retard first 5 detections of every instance (all detections) | R-FCN ResNet-101 · VIDT toy case | average delay (AD): 9.0 -> 13.8 | Hurt |
| tail boost: elevate confidence of late detections | R-FCN ResNet-101 · VIDT toy case | average delay (AD): 9.0 -> 8.9 | No clear effect |
Closest papers by meaning
- Minimum Delay Object Detection From Video (ICCV 2019 · similarity 0.87)
- Searching Parameterized AP Loss for Object Detection (NeurIPS 2021 · similarity 0.86)
- Efficient One-Stage Video Object Detection by Exploiting Temporal Consistency (ECCV 2022 · similarity 0.86)
- Minimum Delay Moving Object Detection (CVPR 2017 · similarity 0.86)
- XS-VID: A Large-Scale Benchmark for Small Object Detection and Tracking in Videos (TPAMI 2026 · similarity 0.85)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · 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.
