Research map / Detection and segmentation / Open-set and few-shot detection
TOIST: Task Oriented Instance Segmentation Transformer with Noun-Pronoun Distillation
Pengfei Li, Beiwen Tian, Yongliang Shi, Xiaoxue Chen, Hao Zhao, Guyue Zhou, Ya-Qin Zhang
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 noun-pronoun distillation (cluster + soft binary target loss) | TOIST · COCO-Tasks | mAP box / mask: 41.3 -> 44.1 box; 35.2 -> 39.0 mask | Improved |
| add self-attention in decoder for preference modeling (ablation) | TOIST · COCO-Tasks | mAP box: 33.9 -> 41.3 | Improved |
| add soft binary target loss (preference distillation) only (ablation) | TOIST · COCO-Tasks | mAP box: 41.3 -> 43.4 | Improved |
| add cluster loss only (ablation) | TOIST · COCO-Tasks | mAP box: 41.3 -> 42.0 | Improved |
| add cluster center replacement only (ablation) | TOIST · COCO-Tasks | mAP box: 41.3 -> 42.0 | Improved |
| distill to prototype via cluster centers instead of directly from noun feature (ablation) | noun-pronoun distillation · COCO-Tasks | mAP box: 41.9 -> 44.1 | Improved |
| use meaningless pronoun abcd instead of something (ablation) | TOIST w/ distillation · COCO-Tasks | mAP box: 44.1 -> 42.8 | Hurt |
| use it or them instead of something (ablation) | TOIST · COCO-Tasks | mAP box: 41.3 -> 41.3 (it), 41.4 (them) | No clear effect |
| add distillation without pre-training (ablation) | TOIST trained from scratch · COCO-Tasks | mAP box: 3.65 -> 7.43 | Improved |
| reduce memory-bank task number n_task from 14 to 1 (ablation) | noun-pronoun distillation · COCO-Tasks, step on something | mAP box: 46.2 -> 47.0 | Improved |
| vary cluster number K from 1 to 10 (ablation) | clustering distillation · COCO-Tasks | mAP box | Improved |
Later papers that compared with this method
1 later accepted paper reports TOIST as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- CoTDet: Affordance Knowledge Prompting for Task Driven Object Detection (ICCV 2023 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.87)
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension (ICLR 2025 · similarity 0.85)
- LISA: Reasoning Segmentation via Large Language Model (CVPR 2024 · similarity 0.85)
- Discriminative Perception via Anchored Description for Reasoning Segmentation (CVPR 2026 · similarity 0.85)
- Prompt-Driven Referring Image Segmentation with Instance Contrasting (CVPR 2024 · similarity 0.85)
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.84)
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.
