Research map / Detection and segmentation / Tracking
Detect to Track and Track to Detect
Christoph Feichtenhofer, Axel Pinz, Andrew Zisserman
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 tracking loss fine-tuning (ablation) | R-FCN ResNet-101 (frame level) · ImageNet VID val | mAP: 74.2 -> 75.8 | Improved |
| link detections into tubes using tracklets and rescore (tau=1) (ablation) | D&T ResNet-101 · ImageNet VID val | mAP: 75.8 -> 79.8 | Improved |
| increase temporal stride from 1 to 10 (ablation) | D&T · ImageNet VID val | mAP: 79.8 -> 78.6 | Hurt |
| use noncausal instead of causal tube rescoring (ablation) | D&T · ImageNet VID val | mAP: 78.7 -> 79.8 | Improved |
| average with detector output at tracked proposal in adjacent frame (tau=1) (ablation) | D&T · ImageNet VID val | mAP: 79.8 -> 80.0 | No clear effect |
| average with detector output at tracked proposal at t+10 (tau=10) (ablation) | D&T · ImageNet VID val | mAP: 78.6 -> 79.2 | Improved |
| use ResNet-101 instead of ResNet-50 backbone | D&T · ImageNet VID val | mAP: 76.5 -> 79.8 | Improved |
| use Inception-v4 instead of ResNet-101 backbone | D&T · ImageNet VID val | mAP: 79.8 -> 82.0 | Improved |
| use ResNeXt-101-32x4 instead of ResNet-101 backbone | D&T · ImageNet VID val | mAP: 79.8 -> 81.4 | Improved |
| tube rescoring on whale class (ablation) | D&T tau=1 · ImageNet VID val | AP whale: -2.6 points | Hurt |
| vary rescoring proportion alpha from 10% to 100% (ablation) | tube rescoring · ImageNet VID val | mAP: less than 0.5% mAP variation | No clear effect |
Closest papers by meaning
- OmniTracker: Unifying Visual Object Tracking by Tracking-With-Detection (TPAMI 2025 · similarity 0.90)
- TrackingNet: A Large-Scale Dataset and Benchmark for Object Tracking in the Wild (ECCV 2018 · similarity 0.90)
- Robust Visual Tracking via Hierarchical Convolutional Features (TPAMI 2018 · similarity 0.90)
- Detect or Track: Towards Cost-Effective Video Object Detection/Tracking (AAAI 2019 · similarity 0.89)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.88)
- DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · similarity 0.88)
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.
