Research map / Detection and segmentation / CNN features
Conditional Convolutions for Instance Segmentation
Zhi Tian, Chunhua Shen, Hao Chen
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 |
|---|---|---|---|
| add relative coordinates to mask head input (ablation) | CondInst · COCO val2017 | mask AP: 31.4 -> 35.7 | Improved |
| use relative coordinates instead of absolute coordinates (ablation) | CondInst · COCO val2017 | mask AP: 32.0 -> 35.7 | Improved |
| add appearance features F_mask to relative coordinates input (ablation) | CondInst · COCO val2017 | mask AP: 31.3 -> 35.7 | Improved |
| upsample mask prediction by factor 2 (ablation) | CondInst · COCO val2017 | mask AP: 34.4 -> 35.8 | Improved |
| upsample mask prediction by 4 rather than 2 (ablation) | CondInst · COCO val2017 | mask AP: 35.8 -> 35.7 | No clear effect |
| deepen mask head from 1 to 3 layers (ablation) | CondInst · COCO val2017 | mask AP: 30.9 -> 35.7 | Improved |
| deepen mask head from 3 to 4 layers (ablation) | CondInst · COCO val2017 | mask AP: 35.7 -> 35.7 | No clear effect |
| widen mask head from 2 to 8 channels (ablation) | CondInst · COCO val2017 | mask AP: 34.1 -> 35.7 | Improved |
| widen mask head from 8 to 16 channels (ablation) | CondInst · COCO val2017 | mask AP: 35.7 -> 35.6 | No clear effect |
| increase C_mask from 1 to 8 (ablation) | CondInst · COCO val2017 | mask AP: 34.8 -> 35.7 | Improved |
| use mask-based NMS instead of box-based NMS (ablation) | CondInst · COCO val2017 | mask AP: 35.7 -> 35.7 | No clear effect |
| add auxiliary semantic segmentation task (ablation) | CondInst R-50 3x · COCO test-dev | mask AP: 37.8 -> 38.8 | Improved |
Closest papers by meaning
- Instance and Panoptic Segmentation Using Conditional Convolutions (TPAMI 2022 · similarity 0.97)
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.90)
- Instance-Sensitive Fully Convolutional Networks (ECCV 2016 · similarity 0.90)
- SOLO: Segmenting Objects by Locations (ECCV 2020 · similarity 0.90)
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.89)
- PointINS: Point-Based Instance Segmentation (TPAMI 2021 · 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.
