Research map / Detection and segmentation / CNN features
Hybrid Task Cascade for Instance Segmentation
Check 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 |
|---|---|---|---|
| Use HTC instead of Cascade Mask R-CNN | ResNet-50-FPN instance segmentation · COCO test-dev | mask AP: 36.9 -> 38.4 | Helped |
| Use HTC instead of Cascade Mask R-CNN | ResNet-101-FPN instance segmentation · COCO test-dev | mask AP: 38.4 -> 39.7 | Helped |
| Use HTC instead of Cascade Mask R-CNN | ResNeXt-101-FPN instance segmentation · COCO test-dev | mask AP: 40.1 -> 41.2 | Helped |
| Interleave bbox and mask branch execution (ablation) | Cascade Mask R-CNN · COCO 2017 val | mask AP: 36.5 -> 36.7 | Helped |
| Add mask information flow between stages (ablation) | Interleaved Cascade Mask R-CNN · COCO 2017 val | mask AP: 36.7 -> 37.4 | Helped |
| Add semantic segmentation branch (ablation) | Cascade with interleaving and mask information flow · COCO 2017 val | mask AP: 37.4 -> 38.0 | Helped |
| Interleave bbox and mask branch execution (ablation) | Mask R-CNN · COCO 2017 val | mask AP: 35.1 -> 35.6 | Helped |
| Add mask information flow between stages (ablation) | Interleaved Cascade Mask R-CNN, ensembled stages · COCO 2017 val | AP: 36.7 -> 37.4 | Helped |
| Add semantic segmentation supervision without feature fusion (ablation) | HTC · COCO 2017 val | AP: 36.5 -> 37.1 | Helped |
| Fuse semantic features with both bbox and mask branches rather than using no fusion (ablation) | HTC with semantic segmentation · COCO 2017 val | AP: 37.1 -> 37.5 | Helped |
| Vary semantic loss weight beta from 0.5 to 3 (ablation) | HTC · COCO 2017 val | AP: 37.9; 38.0; 37.9; 37.8 | No clear effect |
| Add ASPP after FPN (ablation) | HTC · COCO 2017 val | AP: 38.0 -> 38.1 | Helped |
Closest papers by meaning
- Instance-Aware Semantic Segmentation via Multi-Task Network Cascades (CVPR 2016 · similarity 0.92)
- Deeply Shape-Guided Cascade for Instance Segmentation (CVPR 2021 · similarity 0.89)
- RDSNet: A New Deep Architecture forReciprocal Object Detection and Instance Segmentation (AAAI 2020 · similarity 0.88)
- Mask Encoding for Single Shot Instance Segmentation (CVPR 2020 · similarity 0.87)
- Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade (CVPR 2017 · similarity 0.87)
- Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · 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.