Research map / Detection and segmentation / CNN features
An Analysis of Scale Invariance in Object Detection SNIP
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 |
|---|---|---|---|
| Train with SNIP instead of ordinary multi-scale training and testing (ablation) | R-FCN with ResNet-50 · COCO | AP: 35.6 -> 37.8 | Helped |
| Use an image pyramid with SNIP rather than single-scale training and testing (ablation) | R-FCN with ResNet-50 · COCO | AP: 34.5 -> 37.8 | Helped |
| Filter out RoIs outside scale-specific valid ranges during multi-scale training (ablation) | R-FCN with ResNet-50 · COCO small objects | AP_S: 19.5 -> 21.4 | Helped |
| Train at 1400x2000 while ignoring objects larger than 80 pixels (ablation) | R-FCN with ResNet-50 · COCO small objects | AP_S: 19.6 -> 16.4 | Hurt |
| Add SNIP to the improved RPN (ablation) | ResNet-50 RPN · COCO proposals | AR: 61.3 -> 64.0 | Helped |
| Use a stronger DPN-92 backbone for proposal generation (ablation) | COCO RPN · COCO proposals | AR: 64.0 -> 65.7 | Helped |
| Apply SNIP to both RCN and RPN rather than RCN alone (ablation) | D-RFCN with DPN-98 · COCO test-dev | AP: 44.2 -> 44.7 | Helped |
| Add SNIP to the DPN-98 D-RFCN baseline (ablation) | D-RFCN with DPN-98 · COCO test-dev | AP: 41.2 -> 44.2 | Helped |
| Use SNIP rather than standard multi-scale testing (ablation) | R-FCN with ResNet-50 · COCO small objects | AP_S: 19.5 -> 21.4 | Helped |
Closest papers by meaning
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.89)
- Scale-Equalizing Pyramid Convolution for Object Detection (CVPR 2020 · similarity 0.88)
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.88)
- Finding Tiny Faces (CVPR 2017 · similarity 0.88)
- ScaleKD: Distilling Scale-Aware Knowledge in Small Object Detector (CVPR 2023 · similarity 0.88)
- M2Det: A Single-Shot Object Detector Based on Multi-Level Feature Pyramid Network (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.