Text-based Visual Question Answering on TextVQA (Score)
67.32ScoreMRoPE-I
Evaluation Results
| Method | Links | |
|---|---|---|
| MRoPE-IBackbone=Qwen3-VL-8B-Instruct2025.10 | 67.32 | |
| MHRoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 67.2 | |
| MRoPE-I2025.10 | 66.91 | |
| Vanilla RoPE2025.10 | 66.8 | |
| HoPE2025.10 | 66.77 | |
| VideoRoPE2025.10 | 66.58 | |
| MHRoPE2025.10 | 66.49 | |
| MRoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 66.05 | |
| MRoPE2025.10 | 65.85 | |
| CircleRoPE2025.10 | 65.54 | |
| CircleRoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 64.94 | |
| LLaVA-NeXT-13BBackbone=LLaVA-NeXT-13B, Retained Tokens=2880, Pruning Ratio=0%2026.02 | 63.2 | |
| Vanilla RoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 63.15 | |
| LLaVA-1.5Backbone=LLaVA-1.5-13B2026.03 | 63.1 | |
| VisPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=960, Pruning Ratio=66.7%2026.02 | 62.5 | |
| FSRBackbone=LLaVA-NeXT-13B, Retained Tokens=960, Pruning Ratio=66.7%2026.02 | 62.5 | |
| HoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 62.33 | |
| VisionZipBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 62.1 | |
| VisPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 62 | |
| FSRBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 61.9 | |
| CoVFTBackbone=LLaVA-1.5-13B2026.03 | 61.79 | |
| VideoRoPEBackbone=Qwen3-VL-8B-Instruct2025.10 | 61.54 | |
| Full fine-tuningBackbone=LLaVA-1.5-13B2026.03 | 61.36 | |
| CDPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=960, Pruning Ratio=66.7%2026.02 | 61.3 | |
| LLaVA-1.5-13BBackbone=LLaVA-1.5-13B, Retained Tokens=576, Pruning Ratio=0%2026.02 | 61.2 | |
| BitFitBackbone=LLaVA-1.5-13B2026.03 | 61.04 | |
| CDPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 60.9 | |
| FastVBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 60.7 | |
| VisionZipBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 60.7 | |
| HoloVBackbone=LLaVA-NeXT-13B, Retained Tokens=960, Pruning Ratio=66.7%2026.02 | 60.6 | |
| VisPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 60.3 | |
| FSRBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 60.3 | |
| FreezeBackbone=LLaVA-1.5-13B2026.03 | 60.18 | |
| HoloVBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 60 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.02 | 59.7 | |
| Full fine-tuningBackbone=LLaVA-1.5-7B2026.03 | 59.7 | |
| CoVFTBackbone=LLaVA-1.5-7B2026.03 | 59.64 | |
| FSRBackbone=LLaVA-1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.02 | 59.5 | |
| HoloVBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 59.5 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 59.2 | |
| DivPruneBackbone=LLaVA-NeXT-13B, Retained Tokens=640, Pruning Ratio=77.8%2026.02 | 59.2 | |
| LoRABackbone=LLaVA-1.5-7B2026.03 | 59.18 | |
| BitFitBackbone=LLaVA-1.5-7B2026.03 | 59.12 | |
| CDPrunerBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 59 | |
| VisionZipBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58.9 | |
| FSRBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58.8 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.02 | 58.7 | |
| FastVBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58.6 | |
| FreezeBackbone=LLaVA-1.5-7B2026.03 | 58.53 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58.4 | |
| VanillaLoc.=-, Retained Tokens=576, Pruning Ratio=0%2026.03 | 58.2 | |
| LLaVA-1.5Backbone=LLaVA-1.5-7B2026.03 | 58.2 | |
| HoloVBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58.1 | |
| FSRBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 58.1 | |
| HoloVBackbone=LLaVA-1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.02 | 58 | |
| DivPruneBackbone=LLaVA-1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.02 | 58 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 57.9 | |
| SVPTBackbone=LLaVA-1.5-7B2026.03 | 57.89 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 57.6 | |
| DivPruneBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 57.6 | |
| VisionZipBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 57.4 | |
| DivPruneBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 57.4 | |
| DARTLoc.=L2, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 57.4 | |
| HoloVLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 57.4 | |
| VisionZipLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 57.3 | |
| VPTBackbone=LLaVA-1.5-7B2026.03 | 57.29 | |
| HoloVBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 57.1 | |
| VisionZipLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56.8 | |
| HoloVLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56.8 | |
| PRUNESIDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 56.7 | |
| PDropLoc.=L8–L24, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56.6 | |
| PDropLoc.=L8–L24, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 56.5 | |
| DivPruneLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 56.4 | |
| DARTLoc.=L2, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56.4 | |
| SparseVLMLoc.=L1–L32, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 56.1 | |
| DivPruneLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56 | |
| DivPruneLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 55.5 | |
| VisionZipLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 55.5 | |
| FastVBackbone=LLaVA-NeXT-13B, Retained Tokens=320, Pruning Ratio=88.9%2026.02 | 55.4 | |
| HoloVLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 55.4 | |
| AutoSelectLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 55.3 | |
| SparseVLMLoc.=L1–L32, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 54.9 | |
| AutoSelectLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 54.9 | |
| PRUNESIDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 54.7 | |
| DARTLoc.=L2, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 54.4 | |
| AutoSelectLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 54.3 | |
| PRUNESIDLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 54.2 | |
| FastVBackbone=LLaVA-1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.02 | 53.4 | |
| FastVLoc.=L2–L32, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 52.5 | |
| ToMeLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 52.1 | |
| SparseVLMLoc.=L1–L32, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 51.8 | |
| VisNecBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 51.6 | |
| FastVLoc.=L2–L32, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 50.6 | |
| PDropLoc.=L8–L24, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 50.6 | |
| ICONSBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 49.9 | |
| ToMeLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 49.1 | |
| FastVLoc.=L2–L32, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 47.8 | |
| HiREDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 47.4 | |
| HiREDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 46.1 | |
| ToMeLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 45.3 |