Visual Question Answering on VQA v2 (Accuracy and Retention)
78.5AccuracyUpper Bound
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Upper Bound2025.12 | 78.5 | — | |
| FlashVLMToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 76.4 | — | |
| VisPrunerToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 75.8 | — | |
| VisionZipToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 75.6 | — | |
| PruMerge+Token Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 74.7 | — | |
| SparseVLMToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 73.8 | — | |
| FlashVLMToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 73.6 | — | |
| VisPrunerToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 72.7 | — | |
| VisionZipToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 72.4 | — | |
| FlashVLMToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 69.3 | — | |
| SparseVLMToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 68.2 | — | |
| VisPrunerToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 67.7 | — | |
| PruMerge+Token Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 67.4 | — | |
| VisionZipToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 67.1 | — | |
| ToMeToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 63 | — | |
| FastVToken Budget=128 Tokens, Pruning Rate=77.8%2025.12 | 61.8 | — | |
| SparseVLMToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 58.6 | — | |
| ToMeToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 57.1 | — | |
| FastVToken Budget=64 Tokens, Pruning Rate=88.9%2025.12 | 55 | — | |
| PruMerge+Token Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 54.9 | — | |
| ToMeToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 46.8 | — | |
| FastVToken Budget=32 Tokens, Pruning Rate=94.4%2025.12 | 43.4 | — |