Visual Grounding 벤치마크
Visual Grounding on RefCOCO+ testA
Accuracy (%)
- 2021-11-16 — X-VLM (base): Accuracy (%) 89.0
- 2022-11-22 — X2-VLM (large): Accuracy (%) 92.1
- 2023-02-01 — mPLUG-2: Accuracy (%) 92.8
- 2023-11-10 — Florence-2-large-ft: Accuracy (%) 95.3
| Rank | Model | Accuracy (%) | IoU | Extra Training Data | Paper | Code | Year |
|---|---|---|---|---|---|---|---|
| 1 | Florence-2-large-ft | 95.3 | – | ✓ | Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks | retkowsky/florence-2 | 2023 |
| 2 | mPLUG-2 | 92.8 | – | mPLUG-2: A Modularized Multi-modal Foundation Model Across Text, Image and Video | modelscope/modelscope · x-plug/mplug-owl · alibaba/AliceMind · +1 | 2023 | |
| 3 | X2-VLM (large) | 92.1 | – | X$^2$-VLM: All-In-One Pre-trained Model For Vision-Language Tasks | zengyan-97/x-vlm · zengyan-97/x2-vlm | 2022 | |
| 4 | XFM (base) | 90.4 | – | Toward Building General Foundation Models for Language, Vision, and Vision-Language Understanding Tasks | zhangxinsong-nlp/XFM | 2023 | |
| 5 | X2-VLM (base) | 90.3 | – | X$^2$-VLM: All-In-One Pre-trained Model For Vision-Language Tasks | zengyan-97/x-vlm · zengyan-97/x2-vlm | 2022 | |
| 6 | X-VLM (base) | 89.00 | – | Multi-Grained Vision Language Pre-Training: Aligning Texts with Visual Concepts | zengyan-97/x-vlm | 2021 | |
| 7 | HYDRA | – | 61.1 | HYDRA: A Hyper Agent for Dynamic Compositional Visual Reasoning | ControlNet/HYDRA | 2024 |