Fine-grained visual perception on OCRBench (test)
301OCRBench ScoreHoloV + RESTORE
Evaluation Results
| Method | Links | |
|---|---|---|
| HoloV + RESTORERetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 301 | |
| VisPruner + RESTORERetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 298 | |
| VisPrunerRetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 295 | |
| HoloVRetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 295 | |
| HoloV + RESTORERetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 295 | |
| VisPruner + RESTORERetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 294 | |
| DivPrune + RESTORERetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 293 | |
| VisionZip + RESTORERetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 290 | |
| VisPrunerRetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 290 | |
| DivPrune + RESTORERetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 288 | |
| HoloVRetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 288 | |
| VisionZip + RESTORERetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 287 | |
| VisPrunerRetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 287 | |
| VisionZipRetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 286 | |
| VisionZipRetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 285 | |
| HoloV + RESTORERetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 283 | |
| DivPruneRetained Tokens=192, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 281 | |
| HoloVRetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 279 | |
| VisPruner + RESTORERetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 273 | |
| VisionZip + RESTORERetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 269 | |
| DivPrune + RESTORERetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=+ ours2026.06 | 267 | |
| DivPruneRetained Tokens=128, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 264 | |
| VisionZipRetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 258 | |
| DivPruneRetained Tokens=64, Backbone=LLaVA-1.5-7B, Framework=Vanilla2026.06 | 257 |