Update README.md

README.md

@@ -11,18 +11,19 @@ tags:
 - fiftyone
 - image
 - object-detection
-dataset_summary: '
+dataset_summary: >
 
 
 
 
-  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 5308 samples.
+  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 5308
+  samples.
 
 
   ## Installation
 
 
-  If you haven''t already, install FiftyOne:
+  If you haven't already, install FiftyOne:
 
 
   ```bash
@@ -44,9 +45,9 @@ dataset_summary: '
 
   # Load the dataset
 
-  # Note: other available arguments include ''max_samples'', etc
+  # Note: other available arguments include 'max_samples', etc
 
-  dataset = load_from_hub("harpreetsahota/FloorPlanCAD")
+  dataset = load_from_hub("Voxel51/FloorPlanCAD")
 
 
   # Launch the App
@@ -54,16 +55,12 @@ dataset_summary: '
   session = fo.launch_app(dataset)
 
   ```
-
-  '
+license: cc-by-sa-4.0
 ---
 
-# Dataset Card for FloorPlanCAD
-
-<!-- Provide a quick summary of the dataset. -->
-
-
+# Dataset Card for FloorPlanCAD (test split)
 
+![image](floorplancad.gif)
 
 
 This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 5308 samples.
@@ -84,141 +81,232 @@ from fiftyone.utils.huggingface import load_from_hub
 
 # Load the dataset
 # Note: other available arguments include 'max_samples', etc
-dataset = load_from_hub("harpreetsahota/FloorPlanCAD")
+dataset = load_from_hub("Voxel51/FloorPlanCAD")
 
 # Launch the App
 session = fo.launch_app(dataset)
 ```
 
-
 ## Dataset Details
 
 ### Dataset Description
 
-<!-- Provide a longer summary of what this dataset is. -->
-
-
+FloorPlanCAD is a large-scale real-world CAD drawing dataset containing over 15,000 annotated floor plans for panoptic symbol spotting in architectural drawings. The dataset provides line-grained vector annotations for 30 object categories across residential and commercial buildings.
+
+**Key Features:**
+- **Format**: Vector graphics (SVG) with corresponding PNG rasterizations
+- **Scale**: 15,663 CAD drawings (originally 10,094 in v1, updated to 15,663)
+- **Categories**: 30 classes total
+  - 28 "thing" classes (countable instances): doors, windows, furniture, appliances, equipment
+  - 2 "stuff" classes (semantic regions): wall, parking
+- **Annotation Type**: Line-grained primitive-level annotations with semantic and instance labels
+- **Original Split**: 6,382 training / 3,712 testing drawings
+- **Privacy Protected**: Cropped into 20m × 20m blocks, 50% retention rate, sensitive text removed
+
+* **Curated by**: Zhiwen Fan†, Lingjie Zhu†, Honghua Li, Xiaohao Chen, Siyu Zhu, Ping Tan (Alibaba A.I. Labs & Simon Fraser University, †Equal contribution)
+* **Funded by**: Alibaba A.I. Labs
+* **Language(s)**: Not applicable (architectural vector graphics)
+* **License**: Creative Commons Attribution-NonCommercial 4.0 License
+* **Project shutdown notice**: As of January 2023, the project was shut down and most participants left the company
+
+### Dataset Sources
+
+* **Repository**: https://floorplancad.github.io/ (Note: Project shut down in 2022)
+* **Paper**: Fan et al. "FloorPlanCAD: A Large-Scale CAD Drawing Dataset for Panoptic Symbol Spotting" (ICCV 2021)
+  
+## Uses
 
-- **Curated by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Language(s) (NLP):** en
-- **License:** [More Information Needed]
+### Direct Use
 
-### Dataset Sources [optional]
+This dataset is designed for:
+- **Panoptic symbol spotting**: Detecting both countable object instances and semantic regions in architectural drawings
+- **Instance segmentation**: Identifying individual furniture, fixtures, and building elements
+- **Semantic segmentation**: Recognizing structural elements like walls and parking areas
+- **CAD drawing analysis**: Training models for automated floor plan understanding
+- **Architecture/Engineering/Construction (AEC) applications**: Automated 3D modeling from 2D CAD drawings
 
-<!-- Provide the basic links for the dataset. -->
+### Out-of-Scope Use
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+- **Commercial applications**: Dataset is licensed for non-commercial use only
+- **Privacy-sensitive reconstruction**: The dataset is intentionally cropped and anonymized; attempting to reconstruct original complete floor plans or identify building locations violates privacy protections
+- **As-is architectural design**: The cropped 20m × 20m blocks are not complete floor plans suitable for construction
 
-## Uses
+## Dataset Structure
 
-<!-- Address questions around how the dataset is intended to be used. -->
+The converted FiftyOne dataset contains the following structure:
+
+```text
+<Sample: {
+    'id': '690a547c0420c654cb79d521',
+    'media_type': 'image',
+    'filepath': '../image_data/0000-0003.png',
+    'tags': [],
+    'metadata': <ImageMetadata: {
+        'size_bytes': 7803,
+        'mime_type': 'image/png',
+        'width': 1000,
+        'height': 1000,
+        'num_channels': 4,
+    }>,
+    'created_at': datetime.datetime(2025, 11, 4, 19, 31, 8, 427000),
+    'last_modified_at': datetime.datetime(2025, 11, 4, 19, 39, 58, 326000),
+    'ground_truth': <Detections: {
+        'detections': [
+            <Detection: {
+                'id': '690a547c0420c654cb79d520',
+                'attributes': {},
+                'tags': [],
+                'label': 'wall',
+                'bounding_box': [0.30975255, 0.0, 0.69024745, 0.7205705549999999],
+                'mask': array([[255, 255, 255, ...,   0,   0,   0],
+                       [255, 255, 255, ...,   0,   0,   0],
+                       [255, 255, 255, ...,   0,   0,   0],
+                       ...,
+                       [255, 255, 255, ...,   0,   0,   0],
+                       [255, 255, 255, ..., 255, 255, 255],
+                       [255, 255, 255, ..., 255, 255, 255]], dtype=uint8),
+                'mask_path': None,
+                'confidence': None,
+                'index': None,
+            }>,
+        ],
+    }>,
+}>
+```
 
-### Direct Use
+**Object Categories (30 total):**
 
-<!-- This section describes suitable use cases for the dataset. -->
+*Doors (3):* single_door, double_door, sliding_door
 
-[More Information Needed]
+*Windows (4):* window, bay_window, blind_window, opening_symbol
 
-### Out-of-Scope Use
+*Stairs (1):* stair
 
-<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
+*Home Appliances (3):* gas_stove, refrigerator, washing_machine
 
-[More Information Needed]
+*Furniture (11):* sofa, bed, chair, table, bedside_cupboard, tv_cabinet, half_height_cabinet, high_cabinet, wardrobe, sink, bath
 
-## Dataset Structure
+*Equipment (6):* bath_tub, squat_toilet, urinal, toilet, elevator, escalator
 
-<!-- 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. -->
+*Stuff Classes (2):* wall, parking
 
-[More Information Needed]
+**Note on class distribution:** Wall and parking together account for ~27% of all annotated primitives. Significant class imbalance exists across categories.
 
 ## Dataset Creation
 
 ### Curation Rationale
 
-<!-- Motivation for the creation of this dataset. -->
+The FloorPlanCAD dataset was created to address critical limitations in existing symbol spotting research:
 
-[More Information Needed]
+1. **Scale**: Previous datasets (SESYD with 1,000 synthetic plans, FPLAN-POLY with 42 plans) were too small for deep learning
+2. **Real-world diversity**: Prior datasets lacked the symbol variation seen across different architectural firms and building types
+3. **Vector graphics**: Maintaining CAD's native vector format (rather than rasterization) preserves accuracy and enables graph-based methods
+4. **Panoptic scope**: Traditional symbol spotting focused only on "thing" instances; this dataset includes "stuff" classes (walls, parking) for complete scene understanding
 
 ### Source Data
 
-<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
-
 #### 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. -->
+**Original Data Sources:**
+- 100+ architectural projects from production environments
+- Multiple partner companies and institutions
+- Building types: residential towers, schools, hospitals, shopping malls, office buildings
+- Geographic diversity: Projects from various regions (layer names include Chinese characters indicating Asian sources)
 
-[More Information Needed]
+**Technical Processing:**
+- Multi-layer SVG organization (dozens of layers per drawing)
+- Layer-by-layer annotation to reduce clutter
+- Scale handling: Entity lengths range from millimeters to tens of meters (5+ orders of magnitude)
+- Coordinate systems: Metric units (meters) for real-world measurements
 
 #### 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. -->
-
-[More Information Needed]
+- **Primary producers**: Architects, engineers, and CAD designers from various companies creating production floor plans
+- **Data providers**: Multiple partner companies and institutions in the AEC industry (anonymized for privacy)
+- **Geographic origin**: Multinational (layer names suggest significant Asian representation)
+
+### Annotations
+
+#### Annotation Process
+
+**From SVG to FiftyOne Annotations:**
+
+The conversion from vector SVG to structured annotations involves several stages:
+
+1. **SVG Primitive Parsing** (using `svgpathtools`):
+   ```python
+   # For each <path>, <circle>, <ellipse> element:
+   - Extract semantic-id (class label 1-35)
+   - Extract instance-id (unique instance number or -1 for stuff)
+   - Parse geometry:
+     * Paths: start point, end point, middle point via path.point(0.5)
+     * Circles: center (cx, cy), radius (r)
+     * Ellipses: center, radii (rx, ry)
+   ```
+
+2. **Coordinate Transformation**:
+   - SVG coordinates scaled by 10x to match PNG dimensions
+   - svg_x * 10 → png_x
+   - Maintains accurate spatial relationships
+
+3. **Instance Grouping**:
+   - Primitives grouped by (semantic_id, instance_id) tuple
+   - Each unique tuple represents one object instance
+   - "Stuff" classes have instance_id = -1 (no individual instances)
+
+4. **Bounding Box Computation**:
+   ```python
+   # For each instance:
+   - Collect all primitive endpoints and centers
+   - Compute axis-aligned bounding box:
+     x_min = min(all_x_coordinates)
+     y_min = min(all_y_coordinates)
+     width = x_max - x_min
+     height = y_max - y_min
+   - Normalize to [0, 1] by dividing by image dimensions
+   ```
+
+5. **Segmentation Mask Rendering**:
+   ```python
+   # For each instance:
+   - Create blank mask (image_height × image_width)
+   - Render each primitive with line_width=3 pixels:
+     * Paths: cv2.line() or cv2.polylines()
+     * Circles: cv2.circle()
+     * Ellipses: cv2.ellipse()
+   - Crop mask to bounding box region
+   ```
+
+6. **FiftyOne Detection Object Creation**:
+   - Each instance becomes `fo.Detection()` with:
+     * label: mapped class name (e.g., "wall", "single_door")
+     * bounding_box: normalized [x, y, w, h]
+     * mask: binary array (if include_masks=True)
 
-### Annotations [optional]
-
-<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
-
-#### Annotation process
-
-<!-- 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. -->
-
-[More Information Needed]
 
 #### Who are the annotators?
 
-<!-- This section describes the people or systems who created the annotations. -->
-
-[More Information Needed]
-
-#### 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]
-
-## Bias, Risks, and Limitations
-
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
-
-### 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.
-
-## 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. -->
-
-**BibTeX:**
-
-[More Information Needed]
-
-**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]
+- **Number**: 11 specialist annotators
+- **Time investment**: Over 1,000 hours total annotation effort
+- **Expertise**: Domain specialists familiar with architectural CAD drawings
+- **Quality control**: Layer-by-layer annotation methodology for accuracy
+- **Employer**: Alibaba A.I. Labs (annotation team)
+
+## Citation
+
+### BibTeX
+
+```bibtex
+@InProceedings{Fan_2021_ICCV,
+    author    = {Fan, Zhiwen and Zhu, Lingjie and Li, Honghua and Zhu, Siyu and Tan, Ping},
+    title     = {FloorPlanCAD: A Large-Scale CAD Drawing Dataset for Panoptic Symbol Spotting},
+    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
+    month     = {October},
+    year      = {2021},
+    pages     = {10128-10137}
+}
+```
 
-## Dataset Card Contact
+### APA
 
-[More Information Needed]
+Fan, Z., Zhu, L., Li, H., Zhu, S., & Tan, P. (2021). FloorPlanCAD: A Large-Scale CAD Drawing Dataset for Panoptic Symbol Spotting. In *Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)* (pp. 10128-10137).