Research map / Detection and segmentation / Open-set and few-shot detection
GRiT: A Generative Region-to-text Transformer for Object Understanding
Jialian Wu, Jianfeng Wang, Zhengyuan Yang, Zhe Gan, Zicheng Liu, Junsong Yuan, Lijuan Wang
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 |
|---|---|---|---|
| train jointly on detection and dense captioning instead of separately (ablation) | GRiT ViT-B · COCO test-dev | box AP: 53.8 -> 50.7 | Hurt |
| train jointly on detection and dense captioning instead of separately (ablation) | GRiT ViT-B · Visual Genome test | mAP: 15.48 -> 14.37 | Hurt |
| use text decoder (open-set) instead of closed-set classifier (ablation) | GRiT detector · COCO val | box AP: 51.7 -> 50.9 | No clear effect |
| reduce object feature size from 14x14 to 7x7 (ablation) | GRiT · COCO val | box AP: 50.9 -> 50.8 | No clear effect |
| add beam search (size 3) in text decoder (ablation) | GRiT · COCO val | box AP: 50.0 -> 50.9 | Improved |
| increase beam size from 3 to 5 (ablation) | GRiT · COCO val | box AP: 50.9 -> 50.9 | No clear effect |
| add description score to objectness score (beam size 1) (ablation) | GRiT · COCO val | box AP: 49.1 -> 50.0 | Improved |
| add description score to objectness score (beam size 3) (ablation) | GRiT · COCO val | box AP: 19.7 -> 50.9 | Improved |
| add new classes late in training (last one-third) (ablation) | GRiT · COCO val | box AP: 50.9 -> 50.4 | No clear effect |
| use MAE pre-training instead of GIT pre-training (ablation) | GRiT backbone · COCO val | box AP: 52.0 -> 53.6 (ViT-B) | Improved |
| scale backbone from ViT-B to ViT-H (ablation) | GRiT with MAE pre-training · COCO val | box AP: 53.6 -> 58.8 | Improved |
| add Object365 intermediate pre-training | GRiT ViT-H · COCO test-dev | box AP: 59.0 -> 60.4 | Improved |
Closest papers by meaning
- The Devil is in the Fine-Grained Details: Evaluating Open-Vocabulary Object Detectors for Fine-Grained Understanding (CVPR 2024 · similarity 0.88)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.87)
- VisMin: Visual Minimal-Change Understanding (NeurIPS 2024 · similarity 0.87)
- RegionGPT: Towards Region Understanding Vision Language Model (CVPR 2024 · similarity 0.86)
- Improved Visual Grounding through Self-Consistent Explanations (CVPR 2024 · similarity 0.86)
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 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.
