Update README.md

README.md

@@ -9,6 +9,9 @@ task_ids: []
 pretty_name: guiact_smartphone_test
 tags:
 - fiftyone
+- visual-agents
+- os-agents
+- gui-grounding
 - image
 - object-detection
 dataset_summary: '
@@ -46,7 +49,7 @@ dataset_summary: '
 
   # Note: other available arguments include ''max_samples'', etc
 
-  dataset = load_from_hub("harpreetsahota/guiact_smartphone_test")
+  dataset = load_from_hub("Voxel51/guiact_smartphone_test")
 
 
   # Launch the App
@@ -61,7 +64,7 @@ dataset_summary: '
 # Dataset Card for guiact_smartphone_test
 
 <!-- Provide a quick summary of the dataset. -->
-
+![image/png](guiact_smartphone.gif)
 
 
 
@@ -84,141 +87,147 @@ from fiftyone.utils.huggingface import load_from_hub
 
 # Load the dataset
 # Note: other available arguments include 'max_samples', etc
-dataset = load_from_hub("harpreetsahota/guiact_smartphone_test")
+dataset = load_from_hub("Voxel51/guiact_smartphone_test")
 
 # Launch the App
 session = fo.launch_app(dataset)
 ```
 
 
+# GUIAct Smartphone Dataset
+
 ## Dataset Details
 
 ### Dataset Description
 
-<!-- Provide a longer summary of what this dataset is. -->
-
+The GUIAct Smartphone dataset is a collection of multi-step action instructions for smartphone GUI navigation tasks. It was created by converting and adapting a subset of the AITW ([Androids in the Wild](https://arxiv.org/abs/2307.10088)) dataset, specifically focusing on data with the "General" tag. The dataset contains sequences of actions (such as tap, swipe, input) paired with smartphone screenshots, designed to train vision language models to navigate smartphone interfaces. Each instruction requires multiple actions to complete the task, with actions mapped to a standardized format that includes position information. The dataset serves as a crucial component of the larger GUICourse collection for training versatile GUI agents.
 
-
-- **Curated by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
+- **Curated by:** Wentong Chen, Junbo Cui, Jinyi Hu, and other researchers from Tsinghua University, Renmin University of China, and other institutions as listed in the GUICourse paper
+- **Shared by:** The authors of the GUICourse paper
 - **Language(s) (NLP):** en
-- **License:** [More Information Needed]
-
-### Dataset Sources [optional]
+- **License:** CC BY 4.0
 
-<!-- Provide the basic links for the dataset. -->
+### Dataset Sources
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+- **Repository:** https://github.com/yiye3/GUICourse and https://huggingface.co/datasets/yiye2023/GUIAct
+- **Paper:** "GUICourse: From General Vision Language Model to Versatile GUI Agent" (arXiv:2406.11317v1)
 
 ## Uses
 
-<!-- Address questions around how the dataset is intended to be used. -->
-
 ### Direct Use
 
-<!-- This section describes suitable use cases for the dataset. -->
+The GUIAct Smartphone dataset is intended to be used for training and evaluating GUI agents that can navigate smartphone interfaces using visual inputs. Specific use cases include:
 
-[More Information Needed]
+1. Training vision language models to understand smartphone interface elements and their functions
+2. Teaching models to execute multi-step operations on smartphones based on natural language instructions
+3. Developing assistive technologies that can help users navigate smartphone interfaces
+4. Benchmarking the performance of GUI agents on smartphone navigation tasks
 
 ### Out-of-Scope Use
 
-<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
-
-[More Information Needed]
+The dataset is not intended to:
+1. Train models to access private user data on smartphones
+2. Enable unauthorized access to smartphone systems
+3. Replace human decision-making for critical smartphone operations
+4. Train agents to perform tasks beyond the scope of normal smartphone interface interactions
+5. Work with smartphone operating systems or applications not represented in the dataset
 
 ## Dataset Structure
 
-<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
+The GUIAct Smartphone dataset contains 9,157 multi-step action instructions which correspond to approximately 67,000 training samples. Each sample consists of:
 
-[More Information Needed]
+1. A smartphone screenshot (image)
+2. One or more actions to be performed on that screenshot
+3. Natural language instruction describing the task to be accomplished
 
-## Dataset Creation
+The action space includes standardized actions such as:
+- **tap**: Single-point touch on the screen with position coordinates
+- **swipe**: Multi-point gesture with start and end coordinates
+- **input**: Text entry with content specified
+- **enter**: Submission action
+- **answer**: Task completion indication with text output
 
-### Curation Rationale
+Actions include position information represented in either absolute pixel coordinates or relative position format (normalized to a range of 0-1 or 0-1000).
 
-<!-- Motivation for the creation of this dataset. -->
+## FiftyOne Dataset Structure
 
-[More Information Needed]
+# GUIAct Smartphone Test Dataset Structure
 
-### Source Data
+**Basic Info:** 2,079 mobile UI screenshots with interaction annotations
 
-<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
+**Core Fields:**
+- `id`: ObjectIdField - Unique identifier
+- `filepath`: StringField - Image path
+- `tags`: ListField(StringField) - Sample categories
+- `metadata`: EmbeddedDocumentField - Image properties (size, dimensions)
+- `episode`: StringField - Unique identifier for interaction sequence
+- `step`: IntField - Sequential position within episode
+- `question`: StringField - Natural language task description
+- `current_action`: StringField - Current action to perform
+- `structured_history`: ListField - Previous actions in sequence
+- `ui_elements`: EmbeddedDocumentField(Detections) containing multiple Detection objects:
+  - `label`: Element type (ICON_*, TEXT)
+  - `bounding_box`: a list of relative bounding box coordinates in [0, 1] in the following format: `[<top-left-x>, <top-left-y>, <width>, <height>]`
+  - `text`: Text content of element if present
+- `action`: EmbeddedDocumentField(Keypoints) containing interaction points:
+  - `label`: Action type (tap, swipe)
+  - `points`: A list of `(x, y)` keypoints in `[0, 1] x [0, 1]`,  `[[x, y]]` for tap or `[[x1, y1], [x2, y2]]` for swipe
 
-#### Data Collection and Processing
-
-<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
-
-[More Information Needed]
-
-#### Who are the source data producers?
-
-<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
+The dataset captures smartphone screen interactions with detailed UI element annotations and action sequences for mobile interaction research.
 
-[More Information Needed]
-
-### Annotations [optional]
-
-<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
+## Dataset Creation
 
-#### Annotation process
+### Curation Rationale
 
-<!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
+The GUIAct Smartphone dataset was created to provide training data for visual-based GUI agents that can navigate smartphone interfaces. The authors identified limitations in existing datasets for training such agents, including scenarios that were too simplistic, too narrow in domain focus, or too small in size. By adapting data from the larger AITW dataset, they aimed to provide a substantial collection of realistic smartphone navigation tasks for training robust GUI agents.
 
-[More Information Needed]
+### Source Data
 
-#### Who are the annotators?
+#### Data Collection and Processing
 
-<!-- This section describes the people or systems who created the annotations. -->
+The GUIAct Smartphone dataset was derived from the AITW dataset by:
 
-[More Information Needed]
+1. Selecting data with the "General" tag from AITW
+2. Filtering out screenshots that did not include the bottom navigation bar
+3. Converting the original AITW action format to the unified action space defined for GUICourse:
+   - "dual-point gesture" actions were split into "tap" and "swipe" based on distance
+   - "type" actions were renamed to "input" actions
+   - "enter" actions were preserved
+   - "go back/home" actions were converted to "tap" actions on the bottom navigation bar
+   - "task complete/impossible" actions were converted to "answer" actions
 
 #### Personal and Sensitive Information
 
-<!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
-
-[More Information Needed]
+The paper does not explicitly address whether the smartphone screenshots contain personal or sensitive information. However, since the dataset is derived from AITW and intended for research purposes, it likely avoids including personally identifiable information in the screenshots.
 
 ## Bias, Risks, and Limitations
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
+- The dataset may reflect biases present in smartphone interface design
+- The tasks represented may not cover the full diversity of smartphone usage patterns across different user demographics
+- The dataset is likely limited to specific smartphone operating systems and applications
+- Performance of models trained on this data may not generalize to significantly different smartphone interfaces or to future interface designs
+- The action space simplification (e.g., converting dual-point gestures to tap/swipe) may not capture the full complexity of human smartphone interaction
 
 ### Recommendations
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
+Users should be aware that:
+- Models trained on this dataset may perform differently across different smartphone operating systems or UI designs
+- The simplified action space may limit the range of interactions possible with trained models
+- Evaluation should consider both action accuracy and overall task completion
+- For deployment in assistive technologies, additional safety measures and user feedback mechanisms are advisable
+- The dataset should be used as part of a broader training approach that includes ethical considerations for AI assistance
 
-## Citation [optional]
-
-<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
+## Citation
 
 **BibTeX:**
-
-[More Information Needed]
+```bibtex
+@article{chen2024guicourse,
+  title={GUICourse: From General Vision Language Model to Versatile GUI Agent},
+  author={Chen, Wentong and Cui, Junbo and Hu, Jinyi and Qin, Yujia and Fang, Junjie and Zhao, Yue and Wang, Chongyi and Liu, Jun and Chen, Guirong and Huo, Yupeng and Yao, Yuan and Lin, Yankai and Liu, Zhiyuan and Sun, Maosong},
+  journal={arXiv preprint arXiv:2406.11317},
+  year={2024}
+}
+```
 
 **APA:**
-
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
-
-[More Information Needed]
-
-## More Information [optional]
-
-[More Information Needed]
-
-## Dataset Card Authors [optional]
-
-[More Information Needed]
-
-## Dataset Card Contact
-
-[More Information Needed]
+Chen, W., Cui, J., Hu, J., Qin, Y., Fang, J., Zhao, Y., Wang, C., Liu, J., Chen, G., Huo, Y., Yao, Y., Lin, Y., Liu, Z., & Sun, M. (2024). GUICourse: From General Vision Language Model to Versatile GUI Agent. arXiv preprint arXiv:2406.11317.