Visual Question Answering on CVBench 3D
87AccuracyQwen2.5-VL-72B
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-72BParameters=72B2025.07 | 87 | |
| FastVLMLLM Decoder=Qwen2-7B, Data (M) (PT+IT)=15+11.9, Input Res.=1024, #Visual Tokens=256, Latency Enc. (ms)=116.32024.12 | 80.9 | |
| FastVLMLLM Decoder=Qwen2-7B, Data (M) (PT+IT)=15+11.9, Input Res.=768, #Visual Tokens=144, Latency Enc. (ms)=54.82024.12 | 79.3 | |
| TreeVGR-7BParameters=7B2025.07 | 77.6 | |
| Qwen2.5-VL-7BParameters=7B2025.07 | 72.6 | |
| Cambrian-1LLM Decoder=LLama3-8B, Data (M) (PT+IT)=2.5+7, Input Res.=Mult., #Visual Tokens=576, Latency Enc. (ms)=3861.42024.12 | 72 | |
| FastVLMLLM Decoder=Qwen2-7B, Data (M) (PT+IT)=15+6.5, Input Res.=1024, #Visual Tokens=256, Latency Enc. (ms)=116.32024.12 | 69.6 |