Research map / Detection and segmentation / CNN features
Few-Shot Video Object Detection
Qi Fan, Chi-Keung Tang, Yu-Wing Tai
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 |
|---|---|---|---|
| replace RPN with Tube Proposal Network (TPN) (ablation) | matching network (MN) · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 19.5 -> 23.2 | Improved |
| replace image-level MN with temporal matching network (TMN) (ablation) | TPN · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 23.2 -> 26.4 | Improved |
| add label-smoothing regularization to TMN (ablation) | TPN + TMN · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 26.4 -> 27.9 | Improved |
| add support classification module to TMN (ablation) | TPN + TMN · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 26.4 -> 29.4 | Improved |
| combine LSR and SCM (TMN+) (ablation) | TPN + TMN · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 26.4 -> 30.0 | Improved |
| use tube-based features instead of image-based features at inference (ablation) | TMN+ model · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 25.8 -> 30.0 | Improved |
| add relation-network matching vs matching network on RPN (ablation) | RPN · FSVOD-500 val (80 novel classes, full-way 5-shot) | AP: 10.1 -> 19.5 | Improved |
| train on FSVOD-500 (more diverse classes) instead of FSYTV-40 | Ours · FSYTV-40 test | AP: 14.6 -> 17.8 | Improved |
| use joint end-to-end training of video baselines | CenterTrack · FSVOD-500 test | AP: 20.6 -> 22.7 | Improved |
| use joint end-to-end training of video baselines | FairMOT · FSVOD-500 test | AP: 20.3 -> 23.1 | Improved |
| increase number of shots from 1 to 5 in 1-way evaluation | Ours · FSVOD-500 val | AP: 44.0 -> 46.5 | Improved |
| increase number of ways from 1 to 5 in 1-shot evaluation | Ours · FSVOD-500 val | AP: 44.0 -> 31.9 | Hurt |
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.
- FairMOT — FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (IJCV 2021) · better on 3 of 3 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
- Temporal Object-Aware Vision Transformer for Few-Shot Video Object Detection (AAAI 2026 · similarity 0.93)
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.91)
- Few-Shot Object Detection With Attention-RPN and Multi-Relation Detector (CVPR 2020 · similarity 0.91)
- Multi-Scale Positive Sample Refinement for Few-Shot Object Detection (ECCV 2020 · similarity 0.90)
- Query Adaptive Few-Shot Object Detection With Heterogeneous Graph Convolutional Networks (ICCV 2021 · similarity 0.89)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.89)
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.
