Visual Question Answering on SEED (Accuracy)
77AccuracyVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaSource=Qwen’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=0%2026.06 | 77 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 72.7 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 72.1 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 71.7 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 71.1 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 70.1 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 65.1 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 64.9 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 64.9 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 61.8 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 56.8 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 56.6 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 46.9 |