README.md

---
license: apache-2.0
task_categories:
  - object-detection
language:
  - en
tags:
  - flowchart
  - diagram
  - document-analysis
  - object-detection
  - coco-format
size_categories:
  - n<1K
---
<div align="center">
<img src="https://github.com/galirage/gpr-bench/blob/main/assets/galirage_logo.png?raw=true" width="60%" alt="galirage_logo" />

<hr>

[![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)


# FC-Detection
**Flowchart Component Detection Dataset**

🌐 [Galirage Inc.](https://galirage.com) | 📄 [arXiv](https://arxiv.org/abs/2505.07864) | 💻 [GitHub](https://github.com/galirage/Arrow-Guided-VLM-Enhancing-Flowchart-Understanding-via-Arrow-Direction-Encoding)
</div>

# FC-Detection: Flowchart Component Detection Dataset

## Dataset Summary

FC-Detection is a specialized computer vision dataset designed for flowchart component detection and analysis. The dataset contains 99 annotated images of flowcharts with detailed bounding box annotations for 9 different flowchart components, including structural elements (shapes) and directional flow indicators (arrows and endpoints).

This dataset is particularly valuable for document analysis, automated flowchart digitization, workflow understanding, and business process automation applications. The dataset was specifically created and used to train models for the research paper "[Arrow-Guided VLM: Enhancing Flowchart Understanding via Arrow Direction Encoding](https://arxiv.org/abs/2505.07864)" (Omasa et al., 2025).

## Dataset Structure

### Data Splits

The dataset is organized into two splits following COCO annotation format:

| Split | Images | Annotations | Avg. Annotations per Image |
|-------|--------|-------------|---------------------------|
| Train | 69     | 3,951      | ~57.3                    |
| Test  | 30     | 1,875      | ~62.5                    |
| **Total** | **99** | **5,826** | **~58.8** |

### File Structure

```
FC-Detection/
├── data_coco_format_start_end_train/
│   ├── annotations/
│   │   └── instances_custom.json    # COCO format annotations
│   └── images/                      # Training images (69 files)
├── data_coco_format_start_end_test/
│   ├── annotations/
│   │   └── instances_custom.json    # COCO format annotations
│   └── images/                      # Test images (30 files)
└── README.md
```

## Annotation Categories

The dataset detects 9 distinct flowchart components:

| Category ID | Name | Description | Train Annotations | Test Annotations |
|-------------|------|-------------|-------------------|------------------|
| 1 | **text** | Text labels and content within flowchart elements | 967 | 468 |
| 2 | **arrow** | Complete arrow connectors between elements | 765 | 362 |
| 3 | **terminator** | Start/end nodes (oval/rounded rectangles) | 120 | 60 |
| 4 | **data** | Data input/output shapes (parallelograms) | 68 | 22 |
| 5 | **process** | Processing steps (rectangles) | 364 | 182 |
| 6 | **decision** | Decision nodes (diamonds) | 143 | 67 |
| 7 | **connection** | Connection points or junctions | 18 | 6 |
| 8 | **arrow_start** | Arrow starting points/origins | 758 | 356 |
| 9 | **arrow_end** | Arrow ending points/destinations | 748 | 352 |

## Key Features

### Detailed Arrow Endpoint Detection
The dataset provides granular annotation of arrow components:
- **Complete arrows**: Full arrow connectors
- **Arrow start points**: Precise origin locations
- **Arrow end points**: Precise destination locations

This level of detail enables:
- Precise connection analysis between flowchart elements
- Better understanding of flow direction in diagrams
- More accurate relationship mapping between components
- Enhanced arrow tracking in complex flowcharts

### Standard Flowchart Elements
Comprehensive coverage of common flowchart shapes:
- **Terminators**: Start/end points of processes
- **Processes**: Action or processing steps  
- **Decisions**: Conditional branching points
- **Data**: Input/output operations
- **Connections**: Junction points
- **Text**: Labels and descriptions

## Data Format

### Annotation Format
The dataset follows the **COCO (Common Objects in Context)** annotation format:

- **Images**: Metadata including ID, filename, width, height
- **Categories**: 9 predefined flowchart component types
- **Annotations**: Bounding box annotations with:
  - `id`: Unique annotation identifier
  - `image_id`: Reference to parent image
  - `category_id`: Object type (1-9)
  - `bbox`: Bounding box in `[x, y, width, height]` format
  - `area`: Bounding box area in pixels
  - `segmentation`: Empty (bounding box detection only)
  - `iscrowd`: Always 0 (individual objects)

### Image Formats
- **PNG** (primary format)
- **WEBP**
- **JPEG/JPG**
- Variable resolutions from small diagrams to large complex flowcharts

## Usage Example

```python
from datasets import load_dataset
import json
from PIL import Image

# Load the dataset
dataset = load_dataset("galirage/FC-Detection")

# Access train and test splits
train_split = dataset["train"]
test_split = dataset["test"]

# Load COCO annotations manually
with open("data_coco_format_start_end_train/annotations/instances_custom.json", "r") as f:
    train_annotations = json.load(f)

# Categories mapping
categories = {
    1: "text", 2: "arrow", 3: "terminator", 4: "data", 5: "process",
    6: "decision", 7: "connection", 8: "arrow_start", 9: "arrow_end"
}

# Example: Process first training image
first_image = train_annotations["images"][0]
print(f"Image: {first_image['file_name']}, Size: {first_image['width']}x{first_image['height']}")

# Get annotations for this image
image_annotations = [ann for ann in train_annotations["annotations"] if ann["image_id"] == first_image["id"]]
print(f"Number of annotations: {len(image_annotations)}")
```

## Use Cases

### Research Applications
- **Arrow-Guided VLM Research**: Used to train detection models for the "Arrow-Guided VLM" paper, enabling enhanced flowchart understanding through arrow direction encoding
- **Vision-Language Models**: Training object detection components for VLM-based flowchart interpretation systems
- **Flowchart Topology Understanding**: Research in understanding directional flow and node relationships in diagrams

### Practical Applications
- **Document Digitization**: Automatic conversion of hand-drawn or scanned flowcharts
- **Process Mining**: Analysis of business process flows
- **Educational Tools**: Automated assessment of flowchart assignments
- **Workflow Management**: Integration with business process management systems
- **AI Training**: Training object detection models for diagram analysis
- **Research**: Computer vision research in document understanding

## Dataset Statistics

- **Total Images**: 99
- **Total Annotations**: 5,826
- **Average Annotations per Image**: ~58.8
- **Most Common Category**: Text (1,435 total annotations)
- **Least Common Category**: Connection (24 total annotations)
- **Dataset Size**: ~6.78 MB

## Licensing and Citation

This dataset is released under the **Apache 2.0 License**.

## Related Research

This dataset was created to support the research paper:

**"Arrow-Guided VLM: Enhancing Flowchart Understanding via Arrow Direction Encoding"**
*Authors: Takamitsu Omasa, Ryo Koshihara, Masumi Morishige*
*arXiv:2505.07864 (2025)*

---

**Dataset Created by**: Galirage Inc.  
**Contact**: For questions or issues, please open an issue on the dataset repository.  
**Research Paper**: https://arxiv.org/abs/2505.07864