Visual Question Answering on IconVQA
51.5Top-1 AccuracyMiniGPT-v2 (7B)-chat
Evaluation Results
| Method | Links | |
|---|---|---|
| MiniGPT-v2 (7B)-chatGrounding=true, zero-shot=true2023.10 | 51.5 | |
| InfiMM-HDLLM=Vicuna-13B, In-house data=false2024.03 | 51.3 | |
| Sphinx-2KLLM=LLAMA2-13B, In-house data=false2024.03 | 50.5 | |
| MiniGPT-v2 (7B)Grounding=false, zero-shot=true2023.10 | 47.7 | |
| InstructBLIP (13B)Grounding=false, zero-shot=true2023.10 | 44.8 | |
| LLaVA (13B)Grounding=false, zero-shot=true2023.10 | 43 | |
| BLIP-2 (13B)Grounding=false, zero-shot=true2023.10 | 40.6 | |
| BLIP-2LLM=Vicuna-13B, In-house data=false2024.03 | 40.6 | |
| MiniGPT-4 (13B)Grounding=false, zero-shot=true2023.10 | 37.6 |