Visual Reasoning on GQA (Accuracy)
73.01AccuracyActiveScope
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ActiveScopeModel=Qwen3-VL 4B2026.06 | 73.01 | — | |
| ZoomEyeModel=Qwen3-VL 4B2026.06 | 72.82 | — | |
| RegularModel=Qwen3-VL 4B2026.06 | 72.34 | — | |
| ViCropTraining Approach=Training-free, VLM=LLaVA 1.62026.04 | 64.54 | — | |
| VanillaBackbone=LLaVA-Next-7B, Token Reduction Rate=0.0%2025.05 | 64.2 | — | |
| Base modelTraining Approach=Training-free, VLM=LLaVA 1.62026.04 | 64.14 | — | |
| Entropy-Gradient GroundingTraining Approach=Training-free, VLM=LLaVA 1.62026.04 | 63.97 | — | |
| LLaVA-1.5-13BFLOPs=3.817, Ratio=100%, Retained Tokens=Vanilla2026.03 | 63.25 | — | |
| MoB (with η-prior)Backbone=LLaVA-Next-7B, Objectives=PA VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 62.6 | — | |
| VanillaBackbone=LLaVA-v1.5-13B2025.11 | 62.6 | — | |
| LLaVA-NeXT-7BBase Model=LLaVA-NeXT-7B, Retained Tokens=28802026.03 | 62.5 | 100 | |
| ASAPFLOPs=1.253, Ratio=31.88, Retained Tokens=1922026.03 | 62.43 | — | |
| AgilePrunerBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 62 | 99.64 | |
| Base modelTraining Approach=Training-free, VLM=LLaVA 1.52026.04 | 61.98 | — | |
| DivPruneBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 61.9 | 98.9 | |
| VanillaBackbone=LLaVA-1.5-7B, Token Reduction Rate=0.0%2025.05 | 61.9 | — | |
| SINEPROJECT(PO+PD)Backbone=LLaVA-v1.5-13B2025.11 | 61.9 | — | |
| DARTBackbone=LLaVA-Next-7B, Objectives=VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 61.7 | — | |
| POBackbone=LLaVA-v1.5-13B2025.11 | 61.7 | — | |
| PO+PDBackbone=LLaVA-v1.5-13B2025.11 | 61.5 | — | |
| MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 61.4 | — | |
| VanillaBackbone=LLaVA-v1.5-7B2025.11 | 61.3 | — | |
| SparseVLMBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 61.2 | 98.8 | |
| VisionZipBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 61.2 | 99.8 | |
| MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 61.2 | — | |
| Entropy-Gradient GroundingTraining Approach=Training-free, VLM=LLaVA 1.52026.04 | 61.15 | — | |
| DivPruneBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 61.1 | 97 | |
| KL+PDBackbone=LLaVA-v1.5-13B2025.11 | 61.1 | — | |
| Base modelTraining Approach=Training-free, VLM=Qwen 2.5 VL2026.04 | 61.01 | — | |
| ViCropTraining Approach=Training-free, VLM=LLaVA 1.52026.04 | 60.97 | — | |
| MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 60.9 | — | |
| PruMerge+Base Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 60.8 | 98.3 | |
| SINEPROJECT(PO+PD)Backbone=LLaVA-v1.5-7B2025.11 | 60.8 | — | |
| MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 60.7 | — | |
| PO+PDBackbone=LLaVA-v1.5-7B2025.11 | 60.6 | — | |
| POBackbone=LLaVA-v1.5-7B2025.11 | 60.5 | — | |
| AgilePrunerBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 60.1 | 97.94 | |
| PDropBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 60 | 97 | |
| DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 60 | — | |
| PruMerge+FLOPs=1.253, Ratio=32.83, Retained Tokens=1922026.03 | 59.72 | — | |
| KL+PDBackbone=LLaVA-v1.5-7B2025.11 | 59.5 | — | |
| Entropy-Gradient GroundingTraining Approach=Training-free, VLM=Qwen 2.5 VL2026.04 | 59.49 | — | |
| DivpruneFLOPs=1.253, Ratio=32.83, Retained Tokens=1922026.03 | 59.12 | — | |
| Entropy-Gradient GroundingTraining Approach=Training-free, VLM=InternVL 3.52026.04 | 59.01 | — | |
| MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 59 | — | |
| MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 59 | — | |
| FastVBase Model=LLaVA-NeXT-7B, Retained Tokens=6402026.03 | 58.9 | 95.2 | |
| VisionZipBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 58.9 | 96.4 | |
| DivpruneFLOPs=0.833, Ratio=21.801, Retained Tokens=1282026.03 | 58.82 | — | |
| PruMerge+Base Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 58.8 | 95.7 | |
| ASAPFLOPs=0.797, Ratio=20.901, Retained Tokens=1282026.03 | 58.76 | — | |
| DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 58.7 | — | |
| FastVFLOPs=1.259, Ratio=32.99, Retained Tokens=1922026.03 | 58.67 | — | |
| DUET-VLM (C)Speedup=1.3x, Token Budget=Avg 640 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 58.5 | — | |
| DUET-VLM (C)Speedup=1.4x, Token Budget=Avg 320 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 58.5 | — | |
| Qwen-2.5-VL-7BSpeedup=1.00×, Token Budget=Dynamic Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 58.3 | — | |
| VisionZipSpeedup=1.3x, Token Budget=Avg 640 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 58.3 | — | |
| VisionZipSpeedup=1.4x, Token Budget=Avg 320 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 58.2 | — | |
| MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 58.2 | — | |
| Base modelTraining Approach=Training-free, VLM=InternVL 3.52026.04 | 58.04 | — | |
| DUET-VLM (C)Speedup=1.5x, Token Budget=Avg 160 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 57.9 | — | |
| SparseVLMBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 57.9 | 94.6 | |
| SparseVLMFLOPs=0.833, Ratio=21.801, Retained Tokens=1282026.03 | 57.85 | — | |
| SparseVLMFLOPs=1.253, Ratio=32.83, Retained Tokens=1922026.03 | 57.8 | — | |
| GD+PDBackbone=LLaVA-v1.5-7B2025.11 | 57.7 | — | |
| SparseVLMBackbone=LLaVA-1.5-7B, Objectives=PA, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 57.6 | — | |
| PruMerge+FLOPs=0.833, Ratio=21.801, Retained Tokens=1282026.03 | 57.32 | — | |
| VisionZipSpeedup=1.5x, Token Budget=Avg 160 Tokens, Backbone=Qwen-2.5-VL-7B2026.02 | 57.3 | — | |
| MustDropBackbone=LLaVA-Next-7B, Objectives=PA VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 57.3 | — | |
| FastVFLOPs=0.836, Ratio=21.898, Retained Tokens=1282026.03 | 57.19 | — | |
| GD+PDBackbone=LLaVA-v1.5-13B2025.11 | 57 | — | |
| MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 56.9 | — | |
| FasterVLMBackbone=LLaVA-Next-7B, Objectives=VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 56.9 | — | |
| SparseVLMBackbone=LLaVA-Next-7B, Objectives=PA, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 56.1 | — | |
| SparseVLMBackbone=LLaVA-1.5-7B, Objectives=PA, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 56 | — | |
| DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 55.9 | — | |
| FastVBackbone=LLaVA-Next-7B, Objectives=VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.05 | 55.9 | — | |
| MARS2026.02 | 55.7 | — | |
| LoRArank=642026.02 | 55.1 | — | |
| MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 53.1 | — | |
| AdaLoRA2026.02 | 52.7 | — | |
| FastVBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.05 | 52.7 | — | |
| SparseVLMBackbone=LLaVA-1.5-7B, Objectives=PA, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 52.7 | — | |
| PDropBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 50.4 | 82.5 | |
| SEALTraining Approach=Training-based, VLM=Llava 1.52026.04 | 50.18 | — | |
| FastVBase Model=LLaVA-NeXT-7B, Retained Tokens=3202026.03 | 49.8 | 79.9 | |
| FastVBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.05 | 49.6 | — | |
| FastVBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.05 | 46.1 | — | |
| KLBackbone=LLaVA-v1.5-7B2025.11 | 21.8 | — | |
| KLBackbone=LLaVA-v1.5-13B2025.11 | 21.6 | — | |
| GA+PDBackbone=LLaVA-v1.5-7B2025.11 | 19.8 | — | |
| GDBackbone=LLaVA-v1.5-13B2025.11 | 16.4 | — | |
| GDBackbone=LLaVA-v1.5-7B2025.11 | 8.2 | — | |
| GA+PDBackbone=LLaVA-v1.5-13B2025.11 | 6.8 | — | |
| GABackbone=LLaVA-v1.5-7B2025.11 | 0 | — | |
| GABackbone=LLaVA-v1.5-13B2025.11 | 0 | — |