Visual Question Answering on VQA v2 (Robustness Evaluation)
74.5Accuracy (Clean)None
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NoneRetained Tokens=576 (Upper Bound)2026.01 | 74.5 | 55.8 | 49.4 | |
| PruMergeRetained Tokens=1922026.01 | 74.2 | 55.1 | 46.8 | |
| VisionZIPRetained Tokens=1922026.01 | 73.4 | 55.7 | 46.5 | |
| VisPrunerRetained Tokens=1922026.01 | 73.4 | 56.3 | 46.4 | |
| DivPruneRetained Tokens=1922026.01 | 73.1 | 53.9 | 49 | |
| FlowCutRetained Tokens=1922026.01 | 72.9 | 55.9 | 47.3 | |
| PruMergeRetained Tokens=1282026.01 | 72.5 | 55.8 | 47.8 | |
| VisPrunerRetained Tokens=1282026.01 | 72.2 | 56.3 | 46.8 | |
| DivPruneRetained Tokens=1282026.01 | 72.2 | 54 | 49.2 | |
| VisionZIPRetained Tokens=1282026.01 | 72.1 | 53.8 | 45.1 | |
| DivPruneRetained Tokens=642026.01 | 71.5 | 54.3 | 48.3 | |
| VisPrunerRetained Tokens=642026.01 | 70.9 | 53.7 | 45.2 | |
| PruMergeRetained Tokens=642026.01 | 70.7 | 54.3 | 43.9 | |
| FlowCutRetained Tokens=1282026.01 | 70.2 | 54.7 | 44.9 | |
| VisionZIPRetained Tokens=642026.01 | 69.3 | 53.5 | 43 | |
| DivPruneRetained Tokens=322026.01 | 68.7 | 52.4 | 45.2 | |
| VisionZIPRetained Tokens=322026.01 | 68.4 | 51.3 | 43.1 | |
| PruMergeRetained Tokens=322026.01 | 67.6 | 52.8 | 43.6 | |
| VisPrunerRetained Tokens=322026.01 | 66.9 | 51.5 | 43.8 | |
| FlowCutRetained Tokens=642026.01 | 66.8 | 54.3 | 44.4 | |
| NoneCompressed Method=None, Retained Tokens (Kmodel)=144, Backbone=Qwen2.5-VL2026.01 | 65.5 | 40.3 | 40.2 | |
| PruMergeRetained Tokens=162026.01 | 65.2 | 50.5 | 41.8 | |
| DivPruneRetained Tokens=162026.01 | 63.8 | 50.1 | 42.9 | |
| VisionZIPRetained Tokens=162026.01 | 63.2 | 49.7 | 42.7 | |
| FlowCutRetained Tokens=322026.01 | 62.2 | 52.6 | 44.1 | |
| VisPrunerRetained Tokens=162026.01 | 60.4 | 50.9 | 42.4 | |
| DivPruneCompressed Method=DivPrune, Retained Tokens (Kmodel)=72, Backbone=Qwen2.5-VL2026.01 | 58.4 | 38.9 | 35.1 | |
| VisionZIPCompressed Method=VisionZIP, Retained Tokens (Kmodel)=72, Backbone=Qwen2.5-VL2026.01 | 57.4 | 35.7 | 28.1 | |
| AverageCompressed Method=Average, Retained Tokens (Kmodel)=72, Backbone=Qwen2.5-VL2026.01 | 56.9 | 36.9 | 32.3 | |
| FlowCutRetained Tokens=162026.01 | 55.5 | 50.3 | 44.6 | |
| DivPruneCompressed Method=DivPrune, Retained Tokens (Kmodel)=36, Backbone=Qwen2.5-VL2026.01 | 55.3 | 36.2 | 34.9 | |
| VisPrunerCompressed Method=VisPruner, Retained Tokens (Kmodel)=72, Backbone=Qwen2.5-VL2026.01 | 54.9 | 36 | 33.7 | |
| VisionZIPCompressed Method=VisionZIP, Retained Tokens (Kmodel)=36, Backbone=Qwen2.5-VL2026.01 | 53.6 | 28.4 | 27.5 | |
| AverageCompressed Method=Average, Retained Tokens (Kmodel)=36, Backbone=Qwen2.5-VL2026.01 | 53.1 | 32.6 | 31.4 | |
| VisPrunerCompressed Method=VisPruner, Retained Tokens (Kmodel)=36, Backbone=Qwen2.5-VL2026.01 | 50.5 | 33.1 | 31.7 | |
| NoneRetained Tokens=0 (Blind)2026.01 | 44.3 | 44.3 | 44.3 | |
| NoneCompressed Method=None, Retained Tokens (Kmodel)=0, Backbone=Qwen2.5-VL2026.01 | 17 | 17 | 17 |