HumanRef CoT 45K: Object Referring Expressions with Chain-of-Thought Reasoning

A dataset supporting the Rex-Thinker model for grounded object referring via Chain-of-Thought reasoning. It was created by IDEA-Research and last updated on July 23, 2025. The dataset likely contains 45,000 instances of expressions and candidate objects for step-by-step reasoning.

Use Cases

• Training models for interpretable object referring based on the described step-by-step reasoning process.
• Benchmarking models that can reject unmatched expressions, as mentioned in the description.
• Developing Chain-of-Thought reasoning systems for visual grounding tasks.
• Studying the relationship between natural language expressions and visual object candidates.

Strengths

• Designed to address the specific challenges of interpretability and expression rejection mentioned in the description.
• Created by the research organization IDEA-Research, suggesting a research-oriented origin.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset's specific content and format are described only at a high level on an external page.

Provenance

Source

IDEA-Research

Collection Method

Likely created to support the proposed Rex-Thinker model.

Time Range

null

Freshness

Last updated 2025-07-23 04:01:47; freshness should be verified.

Geography

null