Update README.md

README.md

@@ -1,82 +1,152 @@
 ---
 language:
-- en
+  - en
 tags:
-- computer_use
-- agents
-- grounding
-- multimodal
-- ui-vision
-- GroundCUA
+  - computer_use
+  - agents
+  - grounding
+  - multimodal
+  - ui-vision
+  - GroundCUA
 size_categories:
-- 10K<n<100K
+  - 10K<n<100K
+license: mit
+task_categories:
+  - image-text-to-text
 ---
+
+<p align="center">
+  <img src="assets/logo.png" width="48" alt="GroundCUA Logo" style="vertical-align:middle;margin-right:6px;">
+  <strong>GroundCUA: Grounding Computer Use Agents on Human Demonstrations</strong>
+</p>
+
+<p align="center">
+🌐 <a href="https://groundcua.github.io">Website</a> |
+📑 <a href="https://arxiv.org/abs/2511.07332">Paper</a> |
+🤗 <a href="https://huggingface.co/datasets/ServiceNow/GroundCUA">Dataset</a> |
+🤖 <a href="https://huggingface.co/ServiceNow/GroundNext-7B-V0">Models</a>
+</p>
+
+<p align="center">
+  <img src="assets/groundcua-hq.png" width="100%" alt="GroundCUA Overview">
+</p>
+
 # GroundCUA Dataset
 
-## Overview
+GroundCUA is a large and diverse dataset of real UI screenshots paired with structured annotations for building multimodal computer use agents. It covers **87 software platforms** across productivity tools, browsers, creative tools, communication apps, development environments, and system utilities. GroundCUA is designed for research on GUI grounding, UI perception, and vision-language-action models that interact with computers.
+
+---
+
+## Highlights
+
+- **87 platforms** spanning Windows, macOS, Linux, and cross-platform apps  
+- **Annotated UI elements** with bounding boxes, text, and coarse semantic categories  
+- **SHA-256 file pairing** between screenshots and JSON annotations  
+- **Supports research on GUI grounding, multimodal agents, and UI understanding**  
+- **MIT license** for broad academic and open source use  
 
-GroundCUA is a comprehensive dataset containing annotated UI screenshots and corresponding ground truth data for building computer use agents. The dataset covers 87 different software platforms/applications, providing rich annotations for UI element detection, classification, and understanding.
+---
 
 ## Dataset Structure
 
 ```
 GroundCUA/
-├── data/           # Annotation files (JSON format)
-├── images/         # Screenshot images (PNG format)
-├── README.md
+├── data/              # JSON annotation files
+├── images/            # Screenshot images
+└── README.md
 ```
 
-### Directory Organization
+### Directory Layout
+
+Each platform appears as a directory name inside both `data/` and `images/`.
+
+- `data/PlatformName/` contains annotation JSON files  
+- `images/PlatformName/` contains corresponding PNG screenshots  
+
+Image and annotation files share the same SHA-256 hash.
 
-- **`data/`**: Contains 87 platform-specific directories, each with JSON annotation files
-- **`images/`**: Contains 87 platform-specific directories, each with corresponding PNG screenshot images
+---
 
 ## File Naming Convention
 
-Files are named using SHA-256 hashes, ensuring unique identifiers while maintaining correspondence between annotation and image files:
-- `data/Platform/[hash].json` 
-- `images/Platform/[hash].png`
+Each screenshot has a matching annotation file using the same hash:
+
+- `data/PlatformName/[hash].json`  
+- `images/PlatformName/[hash].png`  
 
-Where `[hash]` is the same SHA-256 hash for corresponding annotation and image pairs.
+This structure ensures:
+
+- Unique identifiers for each screenshot  
+- Easy pairing between images and annotations  
+- Compatibility with pipelines that expect hash-based addressing  
+
+---
 
 ## Annotation Format
 
-Each JSON file contains an array of UI element annotations with the following structure:
+Each annotation file is a list of UI element entries describing visible elements in the screenshot.
 
 ```json
 [
-    {
-        "image_path": "Platform/screenshot_hash.png",
-        "bbox": [x1, y1, x2, y2],
-        "text": "UI element text",
-        "category": "Element category",
-        "id": "unique-identifier"
-    },
-    ...
+  {
+    "image_path": "PlatformName/screenshot_hash.png",
+    "bbox": [x1, y1, x2, y2],
+    "text": "UI element text",
+    "category": "Element category",
+    "id": "unique-id"
+  }
 ]
 ```
 
-### Annotation Fields
+### Field Descriptions
+
+**image_path**  
+Relative path to the screenshot.
+
+**bbox**  
+Bounding box coordinates `[x1, y1, x2, y2]` in pixel space.
+
+**text**  
+Visible text or a short description of the element.
+
+**category**  
+Coarse UI type label. Present only for some elements.
+
+**id**  
+Unique identifier for the annotation entry.
+
+---
 
-- **`image_path`**: Relative path to the corresponding screenshot image
-- **`bbox`**: Bounding box coordinates [x1, y1, x2, y2] defining the element's location
-- **`text`**: Text content or description of the UI element
-- **`category`**: Classification of the UI element type. This is a crude classification and is present for few elements
-- **`id`**: Unique identifier for the annotation
+## UI Element Categories
 
-### UI Element Categories
+Categories are approximate and not guaranteed for all elements. Examples include:
 
-The dataset includes several categories of UI elements (these are not avaialble for the entire dataset and maybe ambiguious sometimes):
+- **Button**  
+- **Menu**  
+- **Input Elements**  
+- **Navigation**  
+- **Sidebar**  
+- **Visual Elements**  
+- **Information Display**  
+- **Others**  
 
-- **`Button`**: Interactive buttons and clickable elements
-- **`Menu`**: Interactive buttons and clickable elements
-- **`Input Elements`**: Text fields, dropdowns, checkboxes, and other input controls
-- **`Navigation`**: Menu items, navigation bars, and navigational elements
-- **`Sidebar`**: Sidebars like scrollbars
-- **`Visual Elements`**: Labels, headers, icons, and visual display components
-- **`Information Display`**: Status bars, information panels, and data display elements
-- **`Others`**: Miscellaneous UI elements that don't fit other categories
+These labels provide light structure for UI grounding tasks but do not form a full ontology.
+
+---
+
+## Example Use Cases
+
+GroundCUA can be used for:
+
+- Training computer use agents to perceive and understand UI layouts  
+- Building GUI grounding modules for VLA agents  
+- Pretraining screen parsing and UI element detectors  
+- Benchmarking OCR, layout analysis, and cross-platform UI parsing  
+- Developing models that map UI regions to natural language or actions  
+
+---
 
 ## License
 
-This dataset is released under the MIT License. See the license header for details.
+GroundCUA is released under the MIT License.  
+Users are responsible for ensuring compliance with all applicable laws and policies.