Visual Question Answering on VQA v2 (Accuracy & Relative Score)
80.1AccuracyVanilla
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VanillaBackbone=LLaVA-NeXT, Number of Visual Tokens=2880, Projector Fine-tuning=No2024.12 | 80.1 | 100 | |
| LLaVA-NeXT Vanilla 7BToken Budget=2880, Backbone=LLaVA-NeXT 7B2025.08 | 80.1 | 100 | |
| LLaVA-1.5 Vanilla 13BTokens Retained=576, Compression Ratio=100%2025.08 | 80 | 100 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=640, Projector Fine-tuning=Yes2024.12 | 79.9 | 99.8 | |
| VisionZip (diamond variant)Token Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 79.9 | 99.8 | |
| MMTokToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 79.31 | 99 | |
| LLaVA-1.5-7BReq. Inst.=100%, Sel. Inst.=665K2025.03 | 79.1 | 100 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=640, Projector Fine-tuning=No2024.12 | 79.1 | 98.8 | |
| VisionZipToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 79.1 | 98.8 | |
| DivPruneToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 78.94 | 98.6 | |
| VisionZip 🔥Tokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 78.6 | 98.3 | |
| LLaVA-1.5 Vanilla 7BVisual Tokens=5762025.08 | 78.5 | 100 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=320, Projector Fine-tuning=Yes2024.12 | 78.4 | 97.9 | |
| VisionZip (diamond variant)Token Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 78.4 | 97.9 | |
| MMTokTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 78.3 | 97.9 | |
| Qwen-VL-ChatReq. Inst.=100%, Sel. Inst.=50M2025.03 | 78.2 | — | |
| VisionZipTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 78.1 | 97.6 | |
| DivPruneTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 77.98 | 97.5 | |
| MMTokToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 77.68 | 97 | |
| VisionZip 🔥Tokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 77.6 | 97 | |
| MMTokTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 77.57 | 97 | |
| VisionZip 🔥Visual Tokens=1922025.08 | 77.4 | 98.6 | |
| MMTokVisual Tokens=1922025.08 | 77.11 | 98.2 | |
| SparseVLMBackbone=LLaVA-NeXT, Number of Visual Tokens=640, Projector Fine-tuning=No2024.12 | 77.1 | 96.3 | |
| SparseVLMToken Budget=640, Backbone=LLaVA-NeXT 7B2025.08 | 77.1 | 96.3 | |
| DivPruneTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 77.1 | 96.4 | |
| DivPruneVisual Tokens=1922025.08 | 76.87 | 97.9 | |
| VisionZipVisual Tokens=1922025.08 | 76.8 | 97.8 | |
| VisionZipTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 76.8 | 96 | |
| DivPruneToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 76.64 | 95.7 | |
| VisionZip 🔥Visual Tokens=1282025.08 | 76.6 | 97.6 | |
| MMTokTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 76.55 | 95.7 | |
| MMTokVisual Tokens=1282025.08 | 76.35 | 97.3 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=320, Projector Fine-tuning=No2024.12 | 76.2 | 95.1 | |
| VisionZipToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 76.2 | 95.1 | |
| EL2NReq. Inst.=100%, Sel. Inst.=93K2025.03 | 76.1 | 93 | |
| COINCIDEReq. Inst.=100%, Sel. Inst.=93K2025.03 | 76.1 | 95.5 | |
| TypiClustReq. Inst.=15%, Sel. Inst.=93K2025.03 | 76 | 96.8 | |
| DivPruneVisual Tokens=1282025.08 | 75.96 | 96.8 | |
| MMTokToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 75.62 | 94.4 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=160, Projector Fine-tuning=Yes2024.12 | 75.6 | 94.4 | |
| SparseVLMVisual Tokens=1922025.08 | 75.6 | 96.3 | |
| VisionZipVisual Tokens=1282025.08 | 75.6 | 96.3 | |
| VisionZip (diamond variant)Token Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 75.6 | 94.4 | |
| RandomReq. Inst.=15%, Sel. Inst.=93K2025.03 | 75.3 | 95.7 | |
| MMTokVisual Tokens=642025.08 | 75.2 | 95.8 | |
| DivPruneTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 75.2 | 94 | |
| VisionZip 🔥Tokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 75.2 | 94 | |
| PreSelReq. Inst.=15%, Sel. Inst.=93K2025.03 | 75 | 97.9 | |
| VisionZip 🔥Visual Tokens=642025.08 | 74.2 | 94.5 | |
| DivPruneVisual Tokens=642025.08 | 74.11 | 94.4 | |
| IFDReq. Inst.=100%, Sel. Inst.=93K2025.03 | 74 | 91.8 | |
| Self-FilterReq. Inst.=100%, Sel. Inst.=93K2025.03 | 74 | 88.8 | |
| DivPruneToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 73.92 | 92.3 | |
| SparseVLMVisual Tokens=1282025.08 | 73.8 | 94 | |
| VisionZipTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 73.7 | 92.1 | |
| SparseVLMBackbone=LLaVA-NeXT, Number of Visual Tokens=320, Projector Fine-tuning=No2024.12 | 73.4 | 91.6 | |
| SparseVLMToken Budget=320, Backbone=LLaVA-NeXT 7B2025.08 | 73.4 | 91.6 | |
| PerplexityReq. Inst.=100%, Sel. Inst.=93K2025.03 | 73.1 | 91.9 | |
| VisionZipVisual Tokens=642025.08 | 72.4 | 92.2 | |
| CLIP-ScoreReq. Inst.=100%, Sel. Inst.=93K2025.03 | 71.7 | 90.3 | |
| VisionZipBackbone=LLaVA-NeXT, Number of Visual Tokens=160, Projector Fine-tuning=No2024.12 | 71.4 | 89.1 | |
| VisionZipToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 71.4 | 89.1 | |
| SparseVLMVisual Tokens=642025.08 | 68.2 | 86.9 | |
| FastVVisual Tokens=1922025.08 | 67.1 | 85.5 | |
| SparseVLMBackbone=LLaVA-NeXT, Number of Visual Tokens=160, Projector Fine-tuning=No2024.12 | 66.3 | 82.8 | |
| SparseVLMToken Budget=160, Backbone=LLaVA-NeXT 7B2025.08 | 66.3 | 82.8 | |
| BLIP-2Req. Inst.=100%2025.03 | 65 | — | |
| FastVVisual Tokens=1282025.08 | 61.8 | 78.7 | |
| IDEFICS-80BReq. Inst.=100%, Sel. Inst.=1M2025.03 | 60 | — | |
| FastVVisual Tokens=642025.08 | 55 | 70.1 |