Visual Question Answering on GQA (Accuracy and Efficiency Metrics)
76.87AccuracyVanilla
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VanillaModel=Qwen2-VL-7B2026.03 | 76.87 | 40.84 | 100 | 907.4 | |
| PinPointModel=Qwen2-VL-7B2026.03 | 76.24 | 19.32 | 47.3 | 555.5 | |
| PinPointModel=LLaVA-NeXT-7B2026.03 | 76.08 | 17.96 | 59.1 | 381.6 | |
| VanillaModel=LLaVA-NeXT-7B2026.03 | 75.98 | 30.39 | 100 | 569.5 | |
| Random samplingModel=LLaVA-NeXT-7B2026.03 | 74.84 | 19.26 | 63.4 | 360.7 | |
| FastVModel=LLaVA-NeXT-7B2026.03 | 74.78 | 19.23 | 65.9 | 386 | |
| SparseVLMModel=LLaVA-NeXT-7B2026.03 | 74.49 | 17.85 | 58.7 | 471.1 | |
| PDropModel=LLaVA-NeXT-7B2026.03 | 74.36 | 20.33 | 66.9 | 392 | |
| ToMeModel=LLaVA-NeXT-7B2026.03 | 72.93 | 19.1 | 62.9 | 315.1 | |
| FastVModel=Qwen2-VL-7B2026.03 | 71.8 | 19.59 | 48 | 532.7 | |
| Random samplingModel=Qwen2-VL-7B2026.03 | 70.53 | 20.12 | 49.3 | 530.4 | |
| PDropModel=Qwen2-VL-7B2026.03 | 55.48 | 20.23 | 49.5 | 500.1 |