Research map / Detection and segmentation / Open-set and few-shot detection
CRAFT Objects From Images
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 |
|---|---|---|---|
| Add a cascade proposal classifier and a class-specific one-vs-rest classifier | Faster R-CNN with unshared features · PASCAL VOC 2007 test | mAP: 71.6 -> 75.7 | Helped |
| Add the cascade proposal classifier and class-specific one-vs-rest classifier | Faster R-CNN with unshared features · PASCAL VOC 2012 test | mAP: 65.5 -> 71.3 | Helped |
| Add the class-specific one-vs-rest cascade classifier | CRAFT with cascade proposals · ILSVRC detection val2 | mAP: 47.0 -> 48.5 | Helped |
| Use a strict positive/negative IoU sampling criterion of 0.7/0.3 (ablation) | CRAFT cascade proposal classifier · ILSVRC DET val2 | recall at 0.5 IoU: 92.13 -> 92.37 | Helped |
| Replace the second-stage FRCN with an RPN-like network using a 512-d representation (ablation) | CRAFT cascade proposal classifier · ILSVRC DET val2 | recall at 0.5 IoU: 92.37 -> 91.83 | Hurt |
| Use thresholded proposal scores to keep fewer proposals (ablation) | CRAFT proposal generator · PASCAL VOC 2007 test | mAP: 72.2 -> 72.5 | Helped |
| Add a second detection pass with the same classifier, without fine-tuning (ablation) | Fast R-CNN · PASCAL VOC 2007 test | mAP: 65.0 -> 65.2 | Helped |
| Add a one-vs-rest classifier, fine-tuning only its classifier weights (ablation) | Fast R-CNN · PASCAL VOC 2007 test | mAP: 65.0 -> 66.3 | Helped |
| Add a one-vs-rest classifier and fine-tune fully connected layers (ablation) | Fast R-CNN · PASCAL VOC 2007 test | mAP: 65.0 -> 68.0 | Helped |
| Fine-tune convolutional layers as well as fully connected and classifier layers (ablation) | Fast R-CNN with one-vs-rest classifier · PASCAL VOC 2007 test | mAP: 68.0 -> 67.7 | Hurt |
| Use one-vs-rest classifiers alone instead of combining them with the multiclass classifier (ablation) | Fast R-CNN · PASCAL VOC 2007 test | mAP: 65.0 -> 46.1 | Hurt |
| Use a stricter NMS threshold of 0.6 (ablation) | CRAFT proposal generator · ILSVRC detection val2 | recall at 0.5 IoU: 92.37 -> 93.61 | Helped |
Closest papers by meaning
- Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · similarity 0.90)
- Object Detection Based on Region Decomposition and Assembly (AAAI 2019 · similarity 0.89)
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.88)
- Weakly Supervised Cascaded Convolutional Networks (CVPR 2017 · similarity 0.87)
- HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection (CVPR 2016 · similarity 0.87)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.87)
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.