NaVLM-PVC: Progressive Visual Compression for Efficient Native-Resolution Encoding in MLLMs

Benchmark Model Rank Results
chart-question-answering-on-chartqaNaVLM-PVC–1:1 Accuracy: 82.8
multimodal-reasoning-on-mathvistaNaVLM-PVC–Accuracy: 64.2
multimodal-reasoning-on-mmmuNaVLM-PVC–Accuracy: 50.2
visual-question-answering-vqa-on-realworldqaNaVLM-PVC–Accuracy: 70.3