Research map / Detection and segmentation / CNN features
SOIT: Segmenting Objects with Instance-Aware Transformers
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 |
|---|---|---|---|
| use relative rather than absolute positional encoding (ablation) | SOIT instance-aware transformer · COCO val2017 | mask AP: 38.4 -> 39.2 | Helped |
| add positional encoding rather than removing it (ablation) | SOIT instance-aware transformer · COCO val2017 | mask AP: 37.9 -> 38.4 | Helped |
| add the mask encoder instead of using only linear projection and layer normalization (ablation) | SOIT mask feature branch · COCO val2017 | mask AP: 37.9 -> 38.4 | Helped |
| increase mask feature channels from 4 to 8 (ablation) | SOIT mask encoder · COCO val2017 | mask AP: 37.6 -> 38.4 | Helped |
| increase mask feature channels from 8 to 16 (ablation) | SOIT mask encoder · COCO val2017 | mask AP: 38.4 -> 38.3 | No clear effect |
| increase instance-aware attention heads from 1 to 4 (ablation) | SOIT instance-aware transformer · COCO val2017 | mask AP: 37.8 -> 38.4 | Helped |
| increase instance-aware attention heads from 4 to 8 (ablation) | SOIT instance-aware transformer · COCO val2017 | mask AP: 38.4 -> 38.3 | No clear effect |
| stack a second mask encoder layer (ablation) | SOIT mask encoder · COCO val2017 | mask AP: 38.4 -> 38.4 | No clear effect |
| enable mask loss on more decoder stages (ablation) | SOIT · COCO val2017 | mask AP: 39.2 -> 42.2 | Helped |
| enable mask loss on all decoder stages rather than one (ablation) | SOIT · COCO val2017 | box AP: 46.8 -> 48.9 | Helped |
| use SOIT rather than QueryInst with ResNet-50 | COCO test-dev · COCO test-dev | mask AP: 40.6 -> 42.5 | Helped |
| use SOIT rather than QueryInst with ResNet-101 | COCO test-dev · COCO test-dev | mask AP: 42.8 -> 43.4 | Helped |
Closest papers by meaning
- FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation (CVPR 2023 · similarity 0.89)
- SOLQ: Segmenting Objects by Learning Queries (NeurIPS 2021 · similarity 0.88)
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.88)
- Learning Equivariant Segmentation with Instance-Unique Querying (NeurIPS 2022 · similarity 0.87)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.87)
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 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.