Research map / Detection and segmentation / Open-set and few-shot detection
LSTD: A Low-Shot Transfer Detector for Object Detection
Read the paperCheck 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 LSTD's SSD-style multi-layer regression and Faster RCNN-style coarse-to-fine classification instead of standard detector architectures | Low-shot object detection · Task 1 ImageNet2015 target, 1 shot | mAP: Faster RCNN: 12.2; SSD: 10.1; LSTD: 16.5 | Helped |
| Add transfer-knowledge regularization (ablation) | LSTD fine-tuning · Task 1 ImageNet2015 target, 1–30 shots | mAP: 16.5 -> 18.1; 21.9 -> 25.0; 34.3 -> 35.6; 41.5 -> 43.3; 52.6 -> 55.0 | Helped |
| Add transfer-knowledge regularization (ablation) | LSTD fine-tuning · Task 2 VOC2007 target, 1–30 shots | mAP: 27.1 -> 31.8; 46.1 -> 50.7; 57.9 -> 60.4; 63.2 -> 65.1; 67.2 -> 69.0 | Helped |
| Add transfer-knowledge regularization (ablation) | LSTD fine-tuning · Task 3 VOC2010 target, 1–30 shots | mAP: 29.3 -> 32.7; 37.2 -> 40.8; 48.1 -> 49.7; 52.1 -> 54.1; 56.4 -> 57.9 | Helped |
| Add background-depression regularization to the TK variant (ablation) | LSTD fine-tuning · Task 1 ImageNet2015 target, 1–30 shots | mAP: 18.1 -> 19.2; 25.0 -> 25.8; 35.6 -> 37.4; 43.3 -> 44.3; 55.0 -> 55.8 | Helped |
| Add background-depression regularization to the TK variant (ablation) | LSTD fine-tuning · Task 2 VOC2007 target, 1–30 shots | mAP: 31.8 -> 34.0; 50.7 -> 51.9; 60.4 -> 60.9; 65.1 -> 65.5; 69.0 -> 69.7 | Helped |
| Add background-depression regularization to the TK variant (ablation) | LSTD fine-tuning · Task 3 VOC2010 target, 1–30 shots | mAP: 32.7 -> 33.6; 40.8 -> 42.5; 49.7 -> 50.9; 54.1 -> 54.5; 57.9 -> 58.3 | Helped |
| Use conv7 instead of conv5-3 for ROI pooling (ablation) | LSTD · Task 1 COCO source | mAP: 24.7 -> 25.2 | No clear effect |
| Use conv7 instead of conv5-3 for ROI pooling (ablation) | LSTD · Task 1 ImageNet2015 target, 1 shot | mAP: 15.9 -> 16.5 | No clear effect |
| Apply background-depression regularization at conv7 instead of conv5-3 (ablation) | LSTD, one-shot target fine-tuning · Task 1 | mAP: 19.2 -> 18.9 | Hurt |
| Apply background-depression regularization at conv7 instead of conv5-3 (ablation) | LSTD, one-shot target fine-tuning · Task 2 | mAP: 34.0 -> 34.5 | Helped |
| Apply background-depression regularization at conv7 instead of conv5-3 (ablation) | LSTD, one-shot target fine-tuning · Task 3 | mAP: 33.6 -> 33.4 | Hurt |
Closest papers by meaning
- Context-Transformer: Tackling Object Confusion for Few-Shot Detection (AAAI 2020 · similarity 0.89)
- Weak-Shot Object Detection Through Mutual Knowledge Transfer (CVPR 2023 · similarity 0.88)
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.86)
- Generalized Few-Shot Object Detection Without Forgetting (CVPR 2021 · similarity 0.86)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.84)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.84)
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.