Research map / Detection and segmentation / Open-set and few-shot detection
UniFS: Universal Few-shot Instance Perception with Point Representations
Sheng Jin, Ruijie Yao, Lumin Xu, Wentao Liu, Chen Qian, Ji Wu, Ping Luo
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 SAPL (1-hop) to L1 point loss (ablation) | UniFS · COCO-UniFS val K=1 | Segmentation AP: 7.2 -> 8.4 | Improved |
| extend SAPL from 1-hop to 2-hop (ablation) | UniFS · COCO-UniFS val K=1 | Segmentation AP: 8.4 -> 8.6 | Improved |
| extend SAPL to 3 or 4 hops (ablation) | UniFS · COCO-UniFS val K=5 | Keypoint AP: 22.1 -> 20.9 | Hurt |
| use L1 loss instead of L2 for point regression (ablation) | UniFS · COCO-UniFS val K=1 | Segmentation AP: 6.1 -> 7.2 | Improved |
| add segmentation and keypoint tasks to joint training (ablation) | UniFS · COCO-UniFS val K=1 | Detection AP: 12.2 -> 12.7 | Improved |
| add keypoint task to Det.+Seg. training (ablation) | UniFS · COCO-UniFS val K=5 | Detection AP: 17.6 -> 18.2 | Improved |
| add segmentation to detection-only training (ablation) | UniFS · COCO-UniFS val K=5 | Detection AP: 17.9 -> 17.6 | No clear effect |
| increase segmentation points from 32 to 64 (ablation) | UniFS · COCO-UniFS val | Detection AP: Seg AP K=5 11.2 -> 11.4; Det AP K=5 17.6 -> 17.0 | Mixed |
| increase point decoder blocks beyond 2 (to 4 or 6) (ablation) | UniFS · COCO-UniFS val K=5 | Detection AP: 18.2 -> 17.2 | Hurt |
| use UniFS single model instead of Mask-RCNN-ft multi-task baseline | multi-task few-shot learner · COCO-UniFS val K=5 | Detection AP: 6.9 -> 18.2 | Improved |
| UniFS versus DCFS on 5-shot segmentation | UniFS · COCO-UniFS val K=5 | Segmentation AP: 13.5 -> 11.5 | 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.
- DeFRCN — DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (ICCV 2021) · 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
- Unifying Visual Perception by Dispersible Points Learning (ECCV 2022 · similarity 0.88)
- Universal Instance Perception As Object Discovery and Retrieval (CVPR 2023 · similarity 0.88)
- Single Domain Generalization for Few-Shot Counting via Universal Representation Matching (CVPR 2025 · similarity 0.87)
- Few-Shot Learning via Embedding Adaptation With Set-to-Set Functions (CVPR 2020 · similarity 0.87)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.85)
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.
