Research map / Detection and segmentation / CNN features
Instances As Queries
Yuxin Fang, Shusheng Yang, Xinggang Wang, Yu Li, Chen Fang, Ying Shan, Bin Feng, Wenyu Liu
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 DynConv mask module (parallel dynamic mask heads) to vanilla query-based mask head (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 37.9 -> 39.8 | Improved |
| add parallel supervision on all stages with DynConv (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 38.8 -> 39.8 | Improved |
| add DynConv without parallel supervision (ablation) | vanilla query-based mask head · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 37.4 -> 38.8 | Improved |
| add parallel supervision to vanilla mask head (ablation) | vanilla query-based mask head · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 37.4 -> 37.9 | Improved |
| add HTC mask flow to vanilla query-based mask head (ablation) | query-based cascade mask head · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 37.9 -> 38.9 mask AP; FPS 11.1 -> 6.0 | Mixed |
| add HTC mask flow on top of DynConv (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 39.8 -> 40.0 mask AP; FPS 10.5 -> 5.4 | No clear effect |
| share query and MSA between detection and segmentation (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 38.1 -> 39.8 | Improved |
| share query and MSA between detection and segmentation (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | box AP: 43.4 -> 44.5 | Improved |
| share only MSA (not query) (ablation) | QueryInst · COCO val2017 (R50-FPN, 100 queries, mask head ablation) | mask AP: 38.1 -> 38.3 | No clear effect |
| use QueryInst instead of HTC | HTC · COCO test-dev (R101-FPN, 36 ep, 480-800 w/ crop) | mask AP: 40.8 -> 42.8 | Improved |
| increase queries from 100 to 300 | QueryInst · COCO test-dev (R50-FPN, 36 ep, 640-800) | mask AP: 40.1 -> 40.6; FPS 10.5 -> 7.0 | Improved |
| use QueryInst on Sparse R-CNN detector for box prediction | Sparse R-CNN · COCO test-dev (R101-FPN, 36 ep, 480-800 w/ crop) | box AP: 46.3 -> 48.1 | Improved |
Closest papers by meaning
- FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation (CVPR 2023 · similarity 0.92)
- SOIT: Segmenting Objects with Instance-Aware Transformers (AAAI 2022 · similarity 0.90)
- Learning Equivariant Segmentation with Instance-Unique Querying (NeurIPS 2022 · similarity 0.90)
- MobileInst: Video Instance Segmentation on the Mobile (AAAI 2024 · similarity 0.87)
- Robust and Consistent Online Video Instance Segmentation via Instance Mask Propagation (AAAI 2025 · similarity 0.87)
- Scaling up Image Segmentation across Data and Tasks (CVPR 2025 · similarity 0.86)
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.
