Visual Question Answering on VQAv2 18
82.4AccuracyMipha-3B+ (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| Mipha-3B+ (Ours)LM=Phi-2 (2.7B), Res.=384, PT=558K, IT=665K2024.07 | 82.4 | |
| Mipha-3BLM=Phi-2 (2.7B), Res.=384, PT=558K, IT=665K2024.07 | 81.3 | |
| MoE-LLaVA-3.6BLM=Phi-2 (2.7B), Res.=384, PT=558k, IT=1.59M2024.07 | 79.9 | |
| TinyLLaVALM=Phi-2 (2.7B), Res.=384, PT=1.2M, IT=665k2024.07 | 79.9 | |
| LLaVA-1.5+ (Ours)LM=Vicuna(7B), Res.=336, PT=558K, IT=665K2024.07 | 79.8 | |
| Bunny-3BLM=Phi-2 (2.7B), Res.=384, PT=2M, IT=695K2024.07 | 79.8 | |
| Imp-v1LM=Phi-2 (2.7B), Res.=384, PT=558K, IT=665K2024.07 | 79.5 | |
| mPLUG-Owl2LM=LLAMA (7B), Res.=448, PT=400M, IT=1.23M2024.07 | 79.4 | |
| LAF-7BLM=Vicuna (7B), Res.=336, PT=558K, IT=665K2024.07 | 79 | |
| Qwen-VLLM=Qwen (7B), Res.=448, PT=1.4B, IT=50M2024.07 | 78.8 | |
| LLaVA-1.5LM=Vicuna (7B), Res.=336, PT=558K, IT=665K2024.07 | 78.5 | |
| Qwen-VL-ChatLM=Qwen (7B), Res.=448, PT=1.4B, IT=50M2024.07 | 78.2 | |
| ShikraLM=Vicuna (13B), Res.=224, PT=600K, IT=5.5M2024.07 | 77.4 | |
| IDEFICS-80BLM=LLaMA (65B), Res.=224, PT=353M, IT=1M2024.07 | 60 | |
| IDEFICS-9BLM=LLaMA (7B), Res.=224, PT=353M, IT=1M2024.07 | 50.9 |