Research map / Detection and segmentation / CNN features
Few-Shot End-to-End Object Detection via Constantly Concentrated Encoding across Heads
Jiawei Ma, Guangxing Han, Shiyuan Huang, Yuncong Yang, Shih-Fu Chang
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 the full CoCo regularization (class encodings, SupCT on heads 2-6, background negatives) (ablation) | Sparse-RCNN finetuning · MS COCO 30-shot | nAP: 16.3 -> 20.6 | Improved |
| add class encodings to learnable proposal vectors (ablation) | Sparse-RCNN finetuning · MS COCO 30-shot | nAP: 16.3 -> 17.3 | Improved |
| add SupCT among object encodings of heads 2 to 6 (ablation) | Sparse-RCNN finetuning · MS COCO 30-shot | nAP: 16.3 -> 19.2 | Improved |
| combine class encodings with SupCT on heads 2-6 (ablation) | CoCo-RCNN · MS COCO 30-shot | nAP: 19.2 -> 20.1 | Improved |
| add background object encodings as negatives in SupCT (ablation) | CoCo-RCNN · MS COCO 30-shot | nAP: 20.1 -> 20.6 | Improved |
| use soft sampling from neighbouring heads instead of hard label assignment across heads (ablation) | CoCo regularization · VOC 10-shot | nAP: 13.0 -> 47.2; 18.5 -> 47.2 | Improved |
| use supervised contrastive loss instead of soft-label distillation across heads (ablation) | CoCo regularization · VOC 10-shot | nAP: 43.1 -> 47.2 | Improved |
| use SupCT over positives from several heads instead of plain contrastive (ablation) | CoCo regularization · VOC 10-shot | nAP: 44.5 -> 47.2 | Improved |
| replace SupCT with IoU-weighted CPE loss (ablation) | CoCo regularization · VOC 10-shot | nAP: 46.0 -> 47.2 | No clear effect |
| sample positives using two neighbouring heads instead of previous-head labels only (ablation) | CoCo regularization · VOC 10-shot | nAP: 43.3 -> 47.2 | Improved |
| use separate projectors per head instead of a shared projector (ablation) | CoCo-RCNN · VOC 10-shot / MS COCO 30-shot | nAP50 / nAP: 64.0 -> 65.5; 19.8 -> 20.6 | Improved |
| add CoCo regularization to large-scale training from scratch | Sparse-RCNN · COCO large-scale | AP: 46.3 -> 47.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FSCE — FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding (CVPR 2021) · better on 4 of 5 shared results
- Sparse-RCNN — Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021) · better on 2 of 2 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
- Few-Shot Object Detection via Classification Refinement and Distractor Retreatment (CVPR 2021 · similarity 0.91)
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.90)
- Rethinking Few-Shot Object Detection on a Multi-Domain Benchmark (ECCV 2022 · similarity 0.90)
- Few-Shot Object Detection with Foundation Models (CVPR 2024 · similarity 0.89)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.89)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 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.
