Research map / Detection and segmentation / CNN features
Object Detection in Videos by High Quality Object Linking
Peng Tang, Chunyu Wang, Xinggang Wang, Wenyu Liu, Wenjun Zeng, Jingdong Wang
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 short tubelet linking to CPN + T-NMS (ablation) | object linking detector · ImageNet VID val | mAP (%): 70.5 -> 74.5 | Improved |
| add object linking over whole video (short tubelet linking) | static detector · ImageNet VID val (fast objects) | mAP (%): 47.2 -> 56.0 | Improved |
| add short tubelet linking | static detector · ImageNet VID val (occluded objects) | mAP (%): 65.4 -> 70.4 | Improved |
| replace per-frame NMS with T-NMS in CPN (ablation) | CPN · ImageNet VID val | mAP (%): 69.3 -> 70.5 | Improved |
| use video segment length 2 instead of longer segments (ablation) | T-NMS tubelet detection · ImageNet VID val | mAP (%): +1.4% mAP | Improved |
| link short tubelets in the same frame instead of Seq-NMS across neighboring frames (ablation) | object linking · ImageNet VID val (fast objects) | mAP (%): 49.4 -> 56.0 | Improved |
| replace Seq-NMS with CPN+Seq-NMS (ablation) | Seq-NMS · ImageNet VID val | mAP (%): 70.9 -> 72.1 | Improved |
| use CPN cuboid proposals instead of union proposals (ablation) | short tubelet linking · ImageNet VID val | mAP (%): 72.6 -> 74.5 | Improved |
| use CPN cuboid proposals instead of union proposals (ablation) | proposal generation · Cuboid recall IoU 0.7 | recall (%): 84.9 -> 90.6 | Improved |
| use strict evaluation protocol with instance IDs (ablation) | tubelet evaluation · ImageNet VID val | mAP (%): 70.5 -> 69.8 | Hurt |
| train on VID+DET mixture with multi-scale | object linking detector · ImageNet VID val | mAP (%): 74.5 -> 80.6 | Improved |
Closest papers by meaning
- Object Detection in Videos With Tubelet Proposal Networks (CVPR 2017 · similarity 0.94)
- Object Detection From Video Tubelets With Convolutional Neural Networks (CVPR 2016 · similarity 0.88)
- Action Tubelet Detector for Spatio-Temporal Action Localization (ICCV 2017 · similarity 0.87)
- Relation Distillation Networks for Video Object Detection (ICCV 2019 · similarity 0.87)
- Few-Shot Video Object Detection (ECCV 2022 · similarity 0.87)
- Flow-Guided Feature Aggregation for Video Object Detection (ICCV 2017 · 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.
