Research map / Detection and segmentation / CNN features
Object Proposal Generation Using Two-Stage Cascade SVMs
Ziming Zhang, Philip H. S. Torr
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 |
|---|---|---|---|
| use l1-norm SVMs without ranking constraints (l1-o/r) in both stages (ablation) | two-stage cascade · VOC2007 recall-proposal | object recall | Improved |
| use l1-o/r in Stage I instead of ranking SVMs (ablation) | Stage I classifier · VOC2007 recall-overlap | AUC | Improved |
| use l2-w/r or l1-o/r in Stage II (ablation) | Stage II classifier · VOC2007 recall-overlap | AUC | Improved |
| increase number of quantized scales K from 36 to 121 to 196 (eta 0.5, 0.67, 0.75) (ablation) | quantization scheme · VOC2007 | AUC at 1000 proposals | Improved |
| increase K (ablation) | quantization scheme · VOC2007 | AUC at 1, 10, 100 proposals | No clear effect |
| increase K (ablation) | quantization scheme · VOC2007 | computational time | Hurt |
| use previously proposed ranking cascade setting instead of best l1-o/r setting (ablation) | cascade SVMs · VOC2007 | recall-overlap AUC | Hurt |
| use gradient features with linear filters instead of Baseline A | object proposal pipeline · VOC2007 | computational time (s): 3.58 -> 0.20 | Improved |
| use cascade SVMs instead of Baseline A | object proposal pipeline · VOC2007 (1000 proposals) | object recall (%): 88.6 -> 93.8 | Improved |
| use cascade SVMs instead of Baseline B | object proposal pipeline · VOC2007 (1000 proposals) | object recall (%): 77.7 -> 93.8 | Improved |
| use cascade SVMs instead of Baseline B (structured learning) | object proposal pipeline · VOC2007 (1000 proposals) | AUC objects: 67.4 -> 64.5 | Hurt |
| use cascade SVMs instead of objectness on unseen categories | object proposal pipeline · VOC2007 (14 unseen categories, 10 proposals) | object recall (%): 41.0 -> 46.4 | Improved |
Closest papers by meaning
- CRAFT Objects From Images (CVPR 2016 · similarity 0.87)
- Toward Scale-Invariance and Position-Sensitive Region Proposal Networks (ECCV 2018 · similarity 0.87)
- Sequential Optimization for Efficient High-Quality Object Proposal Generation (TPAMI 2017 · similarity 0.87)
- Object-Proposal Evaluation Protocol is 'Gameable' (CVPR 2016 · similarity 0.86)
- Should All Proposals Be Treated Equally in Object Detection? (ECCV 2022 · similarity 0.85)
- Proposal-Contrastive Pretraining for Object Detection from Fewer Data (ICLR 2023 · similarity 0.85)
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.
