Research map / Detection and segmentation / Open-set and few-shot detection
T-Rex2: Towards Generic Object Detection via Text-Visual Prompt Synergy
Qing Jiang, Feng Li, Zhaoyang Zeng, Shilong Liu, Tianhe Ren, Lei 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 contrastive alignment between text and visual prompts (ablation) | T-Rex2 Swin-T · LVIS val zero-shot, Visual-G | AP: 30.2 -> 34.9 | Improved |
| add contrastive alignment between text and visual prompts (ablation) | T-Rex2 Swin-T · LVIS val zero-shot, Text | AP: 32.2 -> 34.8 | Improved |
| add contrastive alignment between text and visual prompts (ablation) | T-Rex2 Swin-T · COCO val zero-shot, Visual-G | AP: 38.7 -> 38.8 | No clear effect |
| joint text and visual training instead of visual-only training (ablation) | T-Rex2 Swin-T · COCO val zero-shot, Visual-G | AP: 14.0 -> 38.7 | Improved |
| naive joint training without alignment hurts text prompt (ablation) | T-Rex2 Swin-T · COCO val zero-shot, Text | AP: 46.4 -> 44.4 | Hurt |
| increase number of visual prompts from 1 to 16 (ablation) | T-Rex2 Visual-G · COCO val zero-shot | AP: 29.2 -> 38.8 | Improved |
| increase visual prompts from 16 to 64 (ablation) | T-Rex2 Visual-G · LVIS val zero-shot | AP: 34.9 -> 35.2 | No clear effect |
| add SA-1B self-training visual prompt data (ablation) | T-Rex2-T Visual-I (Box) · COCO val zero-shot | AP: 41.1 -> 56.6 | Improved |
| add SA-1B data to generic visual prompt training (ablation) | T-Rex2-T Visual-G · COCO val zero-shot | AP: 41.1 -> 38.8 | Hurt |
| add pseudo-labeled Bamboo classification data to text prompt training (ablation) | T-Rex2-T Text · LVIS minival zero-shot | AP: 34.9 -> 38.7 | Improved |
| text prompt versus visual prompt on common categories | T-Rex2 Swin-T · COCO val-80 zero-shot | AP: 45.8 vs 38.8 | Improved |
| visual prompt versus text prompt on long-tailed ODinW | T-Rex2 Swin-T · ODinW 35val zero-shot | AP avg: 18.0 -> 23.6 | Improved |
Later papers that compared with this method
6 later accepted papers report T-Rex2 as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Ceptor: Vision-Language Model-Infused Diverse Guidance for Detecting Anything (ECCV 2026 · better on 11 of 12 shared results)
- PET-DINO: Unifying Visual Cues into Grounding DINO with Prompt-Enriched Training (CVPR 2026 · better on 5 of 7 shared results)
- T-Rex-Omni: Integrating Negative Visual Prompt in Generic Object Detection (AAAI 2026 · better on 5 of 5 shared results)
- SAM 3: Segment Anything with Concepts (ICLR 2026 · better on 1 of 1 shared result)
- OpenPanoD: Aligning Multimodal Prompts and Spherical Representations for Open-Vocabulary Panoramic Detection (ECCV 2026 · better on 1 of 1 shared result)
- YOLOE: Real-Time Seeing Anything (ICCV 2025 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DetCLIPv2 — DetCLIP: Dictionary-Enriched Visual-Concept Paralleled Pre-training for Open-world Detection (NeurIPS 2022) · 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
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 2025 · similarity 0.90)
- Just a Few Glances: Open-Set Visual Perception with Image Prompt Paradigm (AAAI 2025 · similarity 0.90)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- CP-DETR: Concept Prompt Guide DETR Toward Stronger Universal Object Detection (AAAI 2025 · similarity 0.89)
- Towards Single-Source Domain Generalized Object Detection via Causal Visual Prompts (NeurIPS 2025 · similarity 0.88)
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.
