Unlocking Dense Metric Depth Estimation in VLMs

Benchmark Model Rank Results
object-counting-on-countbenchDepthVLM-8B#1Accuracy: 98.2
object-hallucination-evaluation-on-popeDepthVLM-4B#3F1: 89.9
visual-question-answering-vqa-on-mmbench-cnDepthVLM-8B#11Accuracy: 82.3
visual-question-answering-vqa-on-mmbench-enDepthVLM-8B#17Accuracy: 84.6
visual-question-answering-vqa-on-blinkDepthVLM-8B#20Accuracy: 64.8
optical-character-recognition-ocr-on-ocrbenchDepthVLM-8B#27Score: 862
vision-language-understanding-on-mmstarDepthVLM-8B#67Accuracy: 63.8