Research map / Detection and segmentation / Weakly supervised detection and segmentation
Group R-CNN for Weakly Semi-Supervised Object Detection With Points
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| group proposals from nearby feature points around each annotated point (ablation) | Group R-CNN · COCO RPN proposals, IoU=0.5 | AR: 80.4 -> 90.6 | Helped |
| increase group size from k=3 to k=6 (ablation) | Group R-CNN · COCO RPN proposals, IoU=0.5 | AR: 90.6 -> 92.5 | Helped |
| increase group size from k=6 to k=9 (ablation) | Group R-CNN · COCO RPN proposals, IoU=0.5 | AR: 92.5 -> 92.3 | No clear effect |
| add relative-coordinate features (ablation) | Group R-CNN · COCO val, fixed point annotations | mAP: 35.7 -> 37.1 | Helped |
| use naive instance-level assignment rather than vanilla assignment (ablation) | Group R-CNN with instance grouping · COCO val, fixed point annotations | mAP: 36.6 -> 34.2 | Hurt |
| add one projection layer to the detached R-CNN feature path (ablation) | Group R-CNN · COCO val, fixed point annotations | mAP: 34.2 -> 35.7 | Helped |
| increase projection layers from one to two or three (ablation) | Group R-CNN · COCO val, fixed point annotations | mAP: 35.7 -> 35.8; 35.7 | No clear effect |
| generate dynamic-convolution parameters using both category embeddings and RoI features (ablation) | Group R-CNN · COCO val, fixed point annotations | mAP: 37.1 -> 39.2 | Helped |
| replace dynamic group convolution with regular convolution (ablation) | Group R-CNN · COCO val, fixed point annotations | mAP: 39.2 -> 38.0 | Hurt |
| use kernel size 3 instead of kernel size 1 (ablation) | Group R-CNN dynamic group convolution · COCO val, fixed point annotations | mAP: 39.2 -> 39.5 | No clear effect |
| use instance-level assignment and instance-aware representations (ablation) | Group R-CNN · COCO crowded scenes, 54k images | mAP: 32.2 -> 35.4 (+3.2) | Helped |
| use instance-level assignment and instance-aware representations (ablation) | Group R-CNN · COCO non-crowded scenes, 52k images | mAP: 46.9 -> 48.5 (+1.6) | Helped |
Closest papers by meaning
- Point-Teaching: Weakly Semi-supervised Object Detection with Point Annotations (AAAI 2023 · similarity 0.92)
- Points As Queries: Weakly Semi-Supervised Object Detection by Points (CVPR 2021 · similarity 0.90)
- Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021 · similarity 0.90)
- You Reap What You Sow: Using Videos to Generate High Precision Object Proposals for Weakly-Supervised Object Detection (CVPR 2019 · similarity 0.89)
- Relational Matching for Weakly Semi-Supervised Oriented Object Detection (CVPR 2024 · similarity 0.89)
- Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection (CVPR 2020 · 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.