Image Question Answering on MME Perception
1,599MME-P ScoreQwen2.5-VL-3B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-3BKV Size=2562026.05 | 1,599 | |
| Cambrian-4B + Gaze attentionKV Size=288+42026.05 | 1,589.8 | |
| Cambrian-8B-CLIP-LKV Size=5762026.05 | 1,585.3 | |
| Cambrian-4BKV Size=5762026.05 | 1,580.5 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=1282026.05 | 1,574.1 | |
| Cambrian-4B + Gaze attentionKV Size=144+42026.05 | 1,561.8 | |
| Cambrian-4B + Gaze attentionKV Size=72+42026.05 | 1,539.3 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=54+42026.05 | 1,537 | |
| Cambrian-4B + TokenCompre.KV Size=1442026.05 | 1,533.6 | |
| LLaVA-Next-LLaMA-3-8BKV Size=7842026.05 | 1,526 | |
| LLaVA-OV-7BKV Size=1962026.05 | 1,522.1 | |
| SPHINX-7B2026.05 | 1,515.8 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=36+42026.05 | 1,515.3 | |
| LLaVA-Next-Qwen2-7BKV Size=7842026.05 | 1,453.3 | |
| LLaVA-OV-7B + HERMESKV Size=1002026.05 | 1,452.2 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=502026.05 | 1,426.5 | |
| Cambrian-8B-DINOv2-LKV Size=5762026.05 | 1,366.7 | |
| BLIP-2-7BKV Size=5762026.05 | 1,333.7 | |
| LLaVA-OV-7B + HERMESKV Size=402026.05 | 1,323.4 | |
| Omni-DiffusionModel Type=Any-to-Any, #Params=7B2026.03 | 1,216.7 | |
| InstructBLIPModel Type=Visual LLM†, #Params=14B2026.03 | 1,212.8 | |
| mPLUG-OwlModel Type=Visual LLM†, #Params=7B2026.03 | 976.34 | |
| LLaVAModel Type=Visual LLM†, #Params=7B2026.03 | 809.6 |