Research map / Architectures and efficiency / CNN design
Instance-Sensitive Fully Convolutional Networks
Jifeng Dai, Kaiming He, Yi Li, Shaoqing Ren, Jian Sun
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 |
|---|---|---|---|
| train on full-size images instead of 224x224 crops (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@10: 37.4 -> 38.9 | Improved |
| train on full-size images instead of 224x224 crops (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@1000: 51.4 -> 52.6 | Improved |
| use fully convolutional instance-sensitive maps instead of sliding fc mask regression (both on 224x224 crops) (ablation) | InstanceFCN vs ~DeepMask · PASCAL VOC 2012 val (segment proposals) | AR@10: 31.2 -> 37.4 | Improved |
| use fully convolutional instance-sensitive maps instead of sliding fc mask regression (both on 224x224 crops) (ablation) | InstanceFCN vs ~DeepMask · PASCAL VOC 2012 val (segment proposals) | AR@1000: 47.0 -> 51.4 | Improved |
| increase relative positions from k=3 to k=5 (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@10: 38.3 -> 38.9 | Improved |
| increase relative positions from k=3 to k=5 (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@1000: 52.1 -> 52.6 | Improved |
| increase relative positions from k=5 to k=7 (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@1000: 52.6 -> 52.7 | No clear effect |
| increase relative positions from k=5 to k=7 (ablation) | InstanceFCN · PASCAL VOC 2012 val (segment proposals) | AR@10: 38.9 -> 38.8 | No clear effect |
| replace MNC proposals with InstanceFCN proposals under MNC classifier | MNC classifier · PASCAL VOC 2012 val (instance segmentation) | mAP@0.7: 41.5 -> 43.0 | Improved |
| replace MNC proposals with InstanceFCN proposals under MNC classifier | MNC classifier · PASCAL VOC 2012 val (instance segmentation) | mAP@0.5: 63.5 -> 61.5 | Hurt |
| use InstanceFCN instead of MNC proposals | MNC · PASCAL VOC 2012 val (segment proposals) | AR@10: 33.4 -> 38.9 | Improved |
| use InstanceFCN instead of MNC proposals | MNC · PASCAL VOC 2012 val (segment proposals) | AR@1000: 53.8 -> 52.6 | Mixed |
Closest papers by meaning
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.94)
- Deep Variational Instance Segmentation (NeurIPS 2020 · similarity 0.90)
- Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · similarity 0.90)
- Conditional Convolutions for Instance Segmentation (ECCV 2020 · similarity 0.90)
- SOLO: Segmenting Objects by Locations (ECCV 2020 · similarity 0.90)
- Weakly Supervised Instance Segmentation Using Class Peak Response (CVPR 2018 · similarity 0.89)
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.
