Research map / Detection and segmentation / CNN features
SNIPER: Efficient Multi-Scale Training
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 negative chip mining using proposals to select likely false-positive regions (ablation) | SNIPER · COCO test-dev | AP: 43.4 to 46.1 | Helped |
| Retain the high-resolution scale in three-scale training (ablation) | SNIPER · COCO test-dev | AP: 43.3 to 46.1 | Helped |
| Add OpenImagesV4 pre-training (ablation) | SNIPER with ResNet-101 · COCO test-dev | AP: 46.1 to 46.8 | Helped |
| Add a segmentation head trained with N+1 way softmax loss (ablation) | SNIPER + OpenImages · COCO test-dev | AP: 46.8 to 47.6 | Helped |
| Use binary softmax instead of N+1 way softmax for the mask head (ablation) | SNIPER + OpenImages · COCO instance segmentation | AP: 38.9 to 41.3 | Helped |
| Use N+1 way softmax instead of binary softmax for the mask head (ablation) | SNIPER + OpenImages · COCO test-dev | AP: 47.1 to 47.6 | Helped |
| Omit negative chip sampling (ablation) | SNIPER · COCO minival | AR: 65.4 to 65.4 | No clear effect |
| Use SNIPER multi-scale training rather than SNIP training | Faster R-CNN, ResNet-101 · COCO test-dev | AP: 44.4 to 46.1 | Helped |
| Replace SNIPER training with SSD | MobileNet-v2 detector · COCO test-dev | AP: 22.1 to 34.1 | Hurt |
Closest papers by meaning
- Training-Time-Friendly Network for Real-Time Object Detection (AAAI 2020 · similarity 0.85)
- An Analysis of Scale Invariance in Object Detection SNIP (CVPR 2018 · similarity 0.85)
- Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection (CVPR 2020 · similarity 0.85)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.85)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.84)
- LookWhere? Efficient Visual Recognition by Learning Where to Look and What to See from Self-Supervision (NeurIPS 2025 · similarity 0.83)
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.