ZICQ
中 Log in / Sign up
Newsroom LLMs & Foundation Models #Hugging Face #Vision-Language Models #GUI #Spatial Reasoning #Benchmark

Hugging Face Releases GUI-Primitives: Benchmark for Diagnosing Spatial Reasoning Failures in Vision-Language GUI Groundi

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 6 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:Hugging Face has introduced GUI-Primitives, a novel benchmark designed to evaluate the spatial reasoning capabilities of vision-language models in graphical user interfaces (GUI). The benchmark consists of 994 contrastive instruction pairs across seven spatial relations (left/right, above/below, containment, alignment, proximity, list ordinal, occlusion), addressing the limitation of existing benchmarks in isolating whether models correctly bind relational language to the target element. Test re


Key Breakthroughs

Hugging Face's GUI-Primitives is a benchmark tool specifically designed to evaluate the spatial reasoning capabilities of vision-language models in GUIs. Its main features include:

  • Multi-dimensional Spatial Relations: Covers seven spatial relations including left/right, above/below, containment, alignment, proximity, list ordinal, and occlusion.
  • Contrastive Instruction Pairs: Each instruction pair fixes the screenshot and anchor while changing the relational expression, causing the target to move between two designated candidates.
  • Strict Evaluation Criteria: Uses point-in-box accuracy to assess model performance under strict conditions.

Technical Highlights

  1. Innovative Benchmark Design: GUI-Primitives effectively separates the model's understanding of relational language from its ability to locate target elements through the design of contrastive instruction pairs.
  2. Multi-model Testing: Tested 19 vision-language models, revealing significant deficiencies in spatial reasoning among existing models.
  3. Open Data and Code: The benchmark data, predictions, and code have been made publicly available, providing valuable resources for researchers and developers.

Industry Impact

The release of GUI-Primitives sets a new standard for evaluating the spatial reasoning capabilities of vision-language models, helping to identify and address issues in GUI localization. This is crucial for developing more intelligent and reliable AI assistants and automation tools.

Developer Recommendations

  • Utilize Benchmark Tools: Developers can use GUI-Primitives to evaluate and optimize the spatial reasoning capabilities of their vision-language models.
  • Focus on Model Deficiencies: Analyze test results to identify specific weaknesses in models regarding certain spatial relations and make targeted improvements.
  • Engage in Community Discussions: Join relevant communities to share experiences and suggestions, collectively advancing the development of vision-language models.

Source: Hugging Face Daily Papers (2026-08-22)

— END —

Tags: #Hugging Face #Vision-Language Models #GUI #Spatial Reasoning #Benchmark

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…