Research map / Detection and segmentation / Open-set and few-shot detection
Obj2Seq: Formatting Objects as Sequences with Class Prompt for Visual Tasks
Zhiyang Chen, Yousong Zhu, Zhaowen Li, Fan Yang, Wei Li, Haixin Wang, Chaoyang Zhao, Liwei Wu, Rui Zhao, Jinqiao Wang, Ming Tang
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 Prompted Visual Indicator (class prompt) (ablation) | Deformable DETR · MS COCO detection, R50 | mAP: 44.5 -> 45.1 | Improved |
| add General Sequence Predictor with separate objectness (ablation) | Obj2Seq with class prompt · MS COCO detection, R50 | mAP: 45.1 -> 45.7 | Improved |
| merge objectness into output sequence instead of separate branch (ablation) | Obj2Seq · MS COCO detection, R50 | mAP: 45.1 -> 45.1 | No clear effect |
| initialize prompt vectors with CLIP text features (ablation) | Obj2Seq · MS COCO detection, R50 | mAP: 45.4 -> 45.7 | Improved |
| fix CLIP prompt vectors during training (ablation) | Obj2Seq · MS COCO detection, R50 | mAP: 45.7 -> 45.2 | Hurt |
| vary retention policy (top-K' categories 10 to 80) (ablation) | Obj2Seq · MS COCO detection, R50 | mAP: 0.2% | No clear effect |
| remove class existence score from final object score (ablation) | Obj2Seq · retention policy shift | mAP | Hurt |
| add iterative box refinement | Obj2Seq · MS COCO detection, R50 | mAP: 45.7 -> 46.7 | Improved |
| add iterative box refinement | Deformable DETR · MS COCO detection, R50 | mAP: 44.5 -> 46.3 | Improved |
| replace MLP head with General Sequence Predictor (ablation) | baseline · MS COCO keypoints, 50 epochs | AP: 57.2 -> 60.1 | Improved |
| train longer, 50 -> 150 epochs | Obj2Seq · MS COCO keypoints | AP: 60.1 -> 65.0 | Improved |
| sequence predictor vs 2-Token head, multi-task (ablation) | Obj2Seq · COCO human pose | AP kps: 58.3 -> 60.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Pix2Seq — Pix2seq: A Language Modeling Framework for Object Detection (ICLR 2022) · better on 2 of 2 shared results
- Deformable DETR — Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021) · better on 1 of 1 shared result
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
- Open-Det: An Efficient Learning Framework for Open-Ended Detection (ICML 2025 · similarity 0.85)
- Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks (CVPR 2024 · similarity 0.85)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.85)
- Just a Few Glances: Open-Set Visual Perception with Image Prompt Paradigm (AAAI 2025 · similarity 0.84)
- CTRL-O: Language-Controllable Object-Centric Visual Representation Learning (CVPR 2025 · similarity 0.84)
- General Object Foundation Model for Images and Videos at Scale (CVPR 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.
