Add VisionReasoner UI dataset with 245 samples

README.md

@@ -19,15 +19,23 @@ task_ids:
 
 This dataset contains user interface (UI) images along with associated annotation prompts and solutions for fine-tuning the VisionReasoner model.
 
+## Dataset Description
+
+- **Size**: 245 samples
+- **Task**: Image Segmentation (Semantic Segmentation)
+- **Language**: English
+- **License**: MIT
+
 ## Structure
 
-- `images/`: Folder containing UI images (`.webp` format).
-- `visionreasoner_dataset.parquet`: A `.parquet` file with metadata such as:
-  - Prompt (`problem`)
-  - Segmentation solution (`solution`)
-  - Image file reference
-  - Image height
-- `demo.py`: A custom dataset loading script using Hugging Face `datasets` library.
+- `images/`: Folder containing UI images (`.webp` format)
+- `visionreasoner_dataset.parquet`: Metadata file containing:
+  - `id`: Unique identifier for each sample
+  - `problem`: Annotation prompt describing the UI element to segment
+  - `solution`: JSON-formatted segmentation solution with bounding boxes and points
+  - `image`: Reference to the image file
+  - `img_height`: Image height in pixels
+  - `img_width`: Image width in pixels
 
 ## Usage
 
@@ -38,3 +46,24 @@ from datasets import load_dataset
 
 dataset = load_dataset("shirve13/Demo", trust_remote_code=True)
 print(dataset["train"][0])
+```
+
+## Dataset Loading Script
+
+The dataset uses a custom loading script (`demo.py`) that:
+- Loads metadata from the parquet file
+- Handles image paths correctly
+- Provides proper dataset features for Hugging Face compatibility
+
+## Citation
+
+If you use this dataset in your research, please cite:
+
+```bibtex
+@dataset{visionreasoner_ui_dataset,
+  title={VisionReasoner UI Dataset},
+  author={shirve13},
+  year={2024},
+  url={https://huggingface.co/datasets/shirve13/Demo}
+}
+```

dataset-card.md

@@ -0,0 +1,123 @@
+---
+annotations_creators:
+- user-generated
+language:
+- en
+language_creators:
+- user-generated
+license:
+- mit
+multilinguality:
+- monolingual
+size_categories:
+- 1K<n<10K
+source_datasets:
+- original
+task_categories:
+- image-segmentation
+task_ids:
+- semantic-segmentation
+paperswithcode_id: null
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: visionreasoner_dataset.parquet
+  default: true
+---
+
+# Dataset Card for VisionReasoner UI Dataset
+
+## Dataset Description
+
+- **Repository:** [https://huggingface.co/datasets/shirve13/Demo](https://huggingface.co/datasets/shirve13/Demo)
+- **Paper:** N/A
+- **Point of Contact:** [shirve13](https://huggingface.co/shirve13)
+
+### Dataset Summary
+
+The VisionReasoner UI Dataset contains user interface (UI) images along with associated annotation prompts and solutions for fine-tuning the VisionReasoner model. This dataset is designed for semantic segmentation tasks on UI elements.
+
+### Supported Tasks and Leaderboards
+
+- **Semantic Segmentation**: The dataset is designed for segmenting UI elements based on natural language prompts.
+
+### Languages
+
+The dataset is in English.
+
+## Dataset Structure
+
+### Data Instances
+
+Each instance contains:
+- `id`: Unique identifier for the sample
+- `problem`: Natural language prompt describing the UI element to segment
+- `solution`: JSON-formatted segmentation solution with bounding boxes and points
+- `image`: UI image in WebP format
+- `img_height`: Image height in pixels
+- `img_width`: Image width in pixels
+
+### Data Fields
+
+- `id` (string): Unique identifier for each sample
+- `problem` (string): Annotation prompt describing the UI element to segment
+- `solution` (string): JSON-formatted segmentation solution
+- `image` (image): UI image file
+- `img_height` (int32): Image height in pixels
+- `img_width` (int32): Image width in pixels
+
+### Data Splits
+
+- Train: 245 samples
+
+## Dataset Creation
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+The dataset was created by collecting UI screenshots and annotating them with segmentation prompts and solutions.
+
+#### Who are the source language producers?
+
+The dataset was created by the author for research purposes.
+
+### Annotations
+
+#### Annotation process
+
+UI elements were manually annotated with bounding boxes and segmentation masks based on natural language descriptions.
+
+#### Who are the annotators?
+
+The annotations were created by the dataset author.
+
+### Personal and Sensitive Information
+
+The dataset contains UI screenshots but does not contain personal or sensitive information.
+
+## Additional Information
+
+### Dataset Curators
+
+The dataset was curated by shirve13.
+
+### Licensing Information
+
+This dataset is licensed under the MIT License.
+
+### Citation Information
+
+```bibtex
+@dataset{visionreasoner_ui_dataset,
+  title={VisionReasoner UI Dataset},
+  author={shirve13},
+  year={2024},
+  url={https://huggingface.co/datasets/shirve13/Demo}
+}
+```
+
+### Contributions
+
+Thanks to the Hugging Face community for providing the platform to share this dataset. 

demo.py

@@ -1,5 +1,5 @@
-import datasets
 import os
+import datasets
 
 class Demo(datasets.GeneratorBasedBuilder):
     def _info(self):
@@ -8,27 +8,40 @@ class Demo(datasets.GeneratorBasedBuilder):
                 "id": datasets.Value("string"),
                 "problem": datasets.Value("string"),
                 "solution": datasets.Value("string"),
-                "image": datasets.Image(),
+                "image": datasets.Image(),  # Enables image previews
                 "img_height": datasets.Value("int32"),
+                "img_width": datasets.Value("int32"),
             }),
         )
 
     def _split_generators(self, dl_manager):
+        # Use the current directory where demo.py is located
+        base_path = os.path.dirname(os.path.abspath(__file__))
+        parquet_path = os.path.join(base_path, "visionreasoner_dataset.parquet")
+        
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"filepath": "visionreasoner_dataset.parquet"},
+                gen_kwargs={"filepath": parquet_path, "base_path": base_path}
             ),
         ]
 
-    def _generate_examples(self, filepath):
+    def _generate_examples(self, filepath, base_path):
         import pandas as pd
         df = pd.read_parquet(filepath)
+        
         for idx, row in df.iterrows():
+            # Handle image path correctly
+            if isinstance(row["image"], dict) and "path" in row["image"]:
+                image_path = os.path.join(base_path, row["image"]["path"])
+            else:
+                image_path = str(row["image"])
+            
             yield idx, {
-                "id": row["id"],
-                "problem": row["problem"],
-                "solution": row["solution"],
-                "image": row["image"]["path"],  # e.g., images/ui_img_10.webp
-                "img_height": row["img_height"],
-            }
+                "id": str(row["id"]),
+                "problem": str(row["problem"]),
+                "solution": str(row["solution"]),
+                "image": image_path,  # Full path to the image
+                "img_height": int(row["img_height"]),
+                "img_width": int(row["img_width"]),
+            }