Skip to content

Dataset · lirang04

Visual grounding benchmark for MLLMs

lirang04/GroundingME

An image-text dataset for testing how accurately MLLMs localize objects.

What's inside

GroundingME is a benchmark for zero-shot object detection and visual grounding, containing images and text descriptions with a volume of 1k to 10k examples. According to the model card, the dataset evaluates the ability of multimodal LLMs to localize objects based on descriptions across multiple dimensions. The materials are presented as CVPR 2026 work.

What you can do with it

  • Testing MLLMs on zero-shot object detection
  • Evaluating visual grounding using text descriptions
  • Comparing multimodal models on the benchmark
Tasks
Zero shot object detection
Size
1K–10K rows
Downloads this month
974
Likes
76

How to load it

from datasets import load_dataset

ds = load_dataset("lirang04/GroundingME")