Research map / Detection and segmentation / Open-set and few-shot detection
Hierarchical Shot Detector
Jiale Cao, Yanwei Pang, Jungong Han, Xuelong Li
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 |
|---|---|---|---|
| replace simultaneous cls/reg with one ROC module (regress, offset, classify) (ablation) | SSD-like baseline · COCO minival | AP: 27.8 -> 30.3 | Improved |
| stack a second ROC module (ablation) | HSD one ROC · COCO minival | AP: 30.3 -> 32.6 | Improved |
| add feature enhanced (FE) module between ROCs (ablation) | HSD two ROCs · COCO minival | AP: 32.6 -> 33.3 | Improved |
| increase input size from 320x320 to 512x512 (ablation) | HSD two ROCs + FE · COCO minival | AP: 33.3 -> 38.5 | Improved |
| use classification IoU threshold 0.6 instead of 0.5 for regressed boxes (ablation) | ROC · COCO minival | AP: 28.0 -> 29.9 | Improved |
| use stricter classification IoU threshold 0.8 instead of 0.6 (ablation) | ROC · COCO minival | AP: 29.9 -> 28.2 | Hurt |
| use IoU threshold 0.4 or 0.6 for baseline instead of 0.5 (ablation) | SSD-like baseline · COCO minival | AP: 27.8 -> 28.4 (0.4); 27.8 -> 27.1 (0.6) | Mixed |
| add feature sampling offsets for classification (ablation) | one ROC · COCO minival | AP: 29.9 -> 30.3 | Improved |
| add feature sampling offsets for classification (ablation) | two ROCs · COCO minival | AP: 31.1 -> 32.6 | Improved |
| use two 1x1 convs instead of one to predict sampling offsets (ablation) | ROC · COCO minival | AP: 30.2 -> 30.3 | No clear effect |
| apply FE module to second ROC instead of first (ablation) | HSD two ROCs + FE · COCO minival | AP: 32.7 -> 33.3 | Improved |
| add non-local context to FE module (ablation) | HSD FE second ROC · COCO minival | AP: 33.0 -> 33.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Cascade RCNN — Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018) · better on 0 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
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.84)
- Hierarchical Context Embedding for Region-based Object Detection (ECCV 2020 · similarity 0.83)
- Multi-Scale Positive Sample Refinement for Few-Shot Object Detection (ECCV 2020 · similarity 0.83)
- Offset Bin Classification Network for Accurate Object Detection (CVPR 2020 · similarity 0.83)
- s-Adaptive Decoupled Prototype for Few-Shot Object Detection (ICCV 2023 · similarity 0.83)
- Fine-Grained Prototypes Distillation for Few-Shot Object Detection (AAAI 2024 · similarity 0.82)
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.
