---
license: mit
task_categories:
- object-detection
- keypoint-detection
tags:
- yolo
- aws
- azure
- stride
- security
- keypoints
- graph-extraction
---

# STRIDE Architecture Flow Detection Dataset (AWS & Azure)

## 📌 Overview

This dataset was created to detect flow arrows in cloud architecture diagrams (AWS and Azure),
supporting automated STRIDE threat modeling.

Unlike the component dataset (nodes), this dataset focuses on identifying communication flows
between architectural elements using bounding boxes and keypoints.

Each arrow is annotated with:

- Bounding box (flow_arrow)
- Two keypoints:
  - tail (source)
  - tip (destination)

The goal is to enable automated graph reconstruction from diagrams.

Total images: 4190
Classes: 1 (flow_arrow)
Keypoints per object: 2 (tail, tip)

---

## 🎯 Purpose

- Detect directional flows between architecture components
- Identify communication paths
- Enable graph extraction from diagrams
- Support automated STRIDE flow-based threat analysis

---

## 🖼 Image Source

The same base images used in the STRIDE Architecture Components dataset were reused.

Images were collected from publicly available architecture diagrams, including:

- AWS reference architectures
- Azure reference architectures
- Cloud solution blog posts
- Technical documentation examples

---

## 🔄 Data Augmentation Strategy

The same augmentation pipeline applied to the components dataset was also used here:

- `_BW` (Black and White conversion)
- `_sharp` (Sharpen filter)
- `_contrast` (Contrast adjustment)
- `_gamma_hi` (High gamma correction)
- `_gamma_lo` (Low gamma correction)
- `_jpeg50` (JPEG compression 50%)
- `_blur1` (Gaussian blur level 1)
- `_noise6` (Noise injection level 6)
- `_degrade80` (Quality degradation 80%)

This improves robustness against:

- Low-quality screenshots
- Compression artifacts
- Printed & scanned diagrams
- Export variations

---

## 🏷 Classes

| ID | Class |
|----|--------|
| 0  | flow_arrow |

---

## 📍 Keypoints Definition

Each arrow contains two ordered keypoints:

| Index | Name |
|--------|------|
| 0 | tail (source) |
| 1 | tip (destination) |

These keypoints define directionality and allow reconstruction of communication graphs.

---

## 📦 Dataset Format

YOLO Pose format:

<class_id> <x_center> <y_center> <width> <height> <kp1_x> <kp1_y> <kp1_v> <kp2_x> <kp2_y> <kp2_v>

Where:
- kp1 = tail
- kp2 = tip
- v = visibility flag (0=not labeled, 1=labeled)

All coordinates are normalized between 0 and 1.

---

## 🏗 Structure

images/
  train/
  val/
  test/

labels/
  train/
  val/
  test/

data.yaml

---

## 🔐 STRIDE Flow Mapping Intent

Flows are critical for STRIDE modeling because they represent:

- Trust boundary crossings
- External communications
- Data transfer paths
- Potential attack vectors

By combining:
- Component detection (nodes)
- Flow detection (edges)

It becomes possible to reconstruct a security-aware architecture graph.

---

## 📚 Tools Used

- Label Studio
- YOLO Pose (Ultralytics)
- Hugging Face Hub
- Python augmentation pipeline

---

## ⚠️ Disclaimer

This dataset is intended for research and educational purposes only.
Original diagram copyrights remain with their respective owners.

---

## 👤 Author

Guilherme Santos
Vision Architecture Analyzer – 2026