Visual Question Answering on GQA (Acc., Relative Performance (%))
64.2AccuracyLLaVA-NeXT Vanilla 7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaVA-NeXT Vanilla 7BToken Budget=2880, Backbone=LLaVA-NeXT 7B2025.08 | 64.2 | 100 | |
| VisionZip (diamond variant)Token Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 62.4 | 97.2 | |
| MMTokToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 62.27 | 97 | |
| DivPruneToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 61.58 | 95.9 | |
| VisionZipToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 61.3 | 95.5 | |
| VisionZip (diamond variant)Token Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 61 | 95 | |
| MMTokToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 60.96 | 95 | |
| SparseVLMToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 60.3 | 93.9 | |
| MMTokToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 60.05 | 93.5 | |
| DivPruneToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 59.63 | 92.9 | |
| VisionZipToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 59.3 | 92.4 | |
| VisionZip (diamond variant)Token Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 58.2 | 90.7 | |
| DivPruneToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 57.79 | 90 | |
| SparseVLMToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 57.7 | 89.9 | |
| VisionZipToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 55.5 | 86.4 | |
| SparseVLMToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 51.2 | 79.8 |