Referring Expression Comprehension on RefCOCO UNC (val)
89.16Precision@0.5 IoUGSVA-Llama2-13B
Evaluation results
| Method | Links | |
|---|---|---|
| GSVA-Llama2-13BBackbone=Llama2-13B, Evaluation=Fine-tuned2023.12 | 89.16 | |
| GSVA-Vicuna-13BBackbone=Vicuna-13B, Evaluation=Fine-tuned2023.12 | 87.71 | |
| GSVA-Llama2-13BBackbone=Llama2-13B, Evaluation=Zero-shot2023.12 | 86.99 | |
| GSVA-Vicuna-7BBackbone=Vicuna-7B, Evaluation=Fine-tuned2023.12 | 86.27 | |
| u-LLaVA-7BStrategy=full-ft, Output Head=mask2bbox2023.12 | 86.04 | |
| LISA-Vicuna-13BBackbone=Vicuna-13B, Evaluation=Fine-tuned2023.12 | 85.92 | |
| LISA-Llama2-13BBackbone=Llama2-13B, Evaluation=Fine-tuned2023.12 | 85.91 | |
| GSVA-Vicuna-7BBackbone=Vicuna-7B, Evaluation=Zero-shot2023.12 | 85.5 | |
| LISA-Vicuna-7BBackbone=Vicuna-7B, Evaluation=Fine-tuned2023.12 | 85.39 | |
| u-LLaVA-7BStrategy=LoRA, Output Head=mask2bbox2023.12 | 83.47 | |
| GSVA-Vicuna-13BBackbone=Vicuna-13B, Evaluation=Zero-shot2023.12 | 83.12 | |
| LISA-Llama2-13BBackbone=Llama2-13B, Evaluation=Zero-shot2023.12 | 82.52 | |
| LISA-Vicuna-13BBackbone=Vicuna-13B, Evaluation=Zero-shot2023.12 | 80.01 | |
| LISA-Vicuna-7BBackbone=Vicuna-7B, Evaluation=Zero-shot2023.12 | 78.68 |