Research map / Detection and segmentation / Open-set and few-shot detection
Not All Labels Are Equal: Rationalizing the Labeling Costs for Training Object Detection
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 |
|---|---|---|---|
| Use the unified entropy-and-inconsistency acquisition score with selective pseudo-labeling instead of MC-dropout acquisition | SSD300 VGG · VOC07+12 test, cycle 1 | mAP@0.5: 70.95 vs 66.90 | Helped |
| Use the proposed method instead of the strongest listed baseline, Ensemble | SSD300 VGG · VOC07+12 test, cycle 5 | mAP@0.5: 75.60 vs 74.29 | Helped |
| Use the proposed method instead of PM | SSD300 VGG · MS-COCO val2017, cycle 1 | mAP@0.5: 30.07 vs 29.28 | Helped |
| Use the proposed method instead of PM | SSD300 VGG · MS-COCO val2017, cycle 5 | mAP@0.5: 32.80 vs 32.27 | Helped |
| Combine entropy and inconsistency rather than using either score alone (ablation) | Active learning for SSD300 VGG · VOC07+12 test, cycle 1 | mAP@0.5: 67.24 -> 68.40; 67.39 -> 68.40 | Helped |
| Add the unified acquisition score to the full method, including pseudo-labeling (ablation) | SSD300 VGG · VOC07+12 test, cycle 1 | mAP@0.5: 68.40 -> 70.95 | Helped |
| Add selective pseudo-labeling to the unified acquisition method (ablation) | SSD300 VGG · MS-COCO val2017, cycle 1 | mAP@0.5: 29.17 -> 30.07 | Helped |
| Remove consistency-based SSL training (ablation) | Inconsistency acquisition for SSD300 VGG · VOC07+12 test, cycle 1 | mAP@0.5: 64.23 -> 63.26 | Hurt |
| Remove consistency-based SSL training (ablation) | Inconsistency acquisition for SSD300 VGG · VOC07+12 test, cycle 5 | mAP@0.5: 69.27 -> 70.33 | Hurt |
| Lower the pseudo-label confidence threshold from 0.99 to 0.9 (ablation) | Pseudo-labeling for SSD300 VGG · VOC07+12 test, cycle 1 | pseudo-label correctness: 96.68 -> 91.13 | Hurt |
| Lower the pseudo-label confidence threshold from 0.99 to 0.5 (ablation) | Pseudo-labeling for SSD300 VGG · VOC07+12 test, cycle 1 | pseudo-label correctness: 96.68 -> 80.04 | Hurt |
| Use class-wise top-k% pseudo-label selection instead of a class-agnostic confidence threshold (ablation) | Pseudo-labeling for SSD300 VGG · VOC07+12 test, cycle 1 | mAP@0.5: 69.42; 69.63; 69.67 vs 70.95 | Hurt |
Closest papers by meaning
- Portable Active Learning for Object Detection (CVPR 2026 · similarity 0.89)
- Towards Human-Machine Cooperation: Self-Supervised Sample Mining for Object Detection (CVPR 2018 · similarity 0.89)
- Box-Level Active Detection (CVPR 2023 · similarity 0.88)
- Plug and Play Active Learning for Object Detection (CVPR 2024 · similarity 0.88)
- Active Domain Adaptation with False Negative Prediction for Object Detection (CVPR 2024 · similarity 0.86)
- LEOD: Label-Efficient Object Detection for Event Cameras (CVPR 2024 · similarity 0.86)
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.