Dataset Setup¶

The framework uses several datasets for wire detection, component detection, and evaluation. These datasets are not included in the repository — they must be downloaded or generated separately.

Available Datasets¶

Key	Source	Images	Labels	Purpose
`hdc`	Roboflow export	1,993	YOLO OBB (57 component classes)	Component detection training + masked wire detection
`hand_drawn`	Roboflow export	140	YOLO OBB (wire polygons)	Wire detection evaluation benchmark
`synthetic`	Self-generated	2,000+	YOLOv8 pose / lines	Large-scale parameter sweeps
`database`	Local collection	662	None	Visual inspection / real-world testing

Quick Start¶

1. Download Datasets¶

HDC-Recognition and Hand-drawn wires are exported from Roboflow. Place them at the repo root:

LineDetection/
├── roboflow_test2/          # HDC dataset (1993 images)
│   ├── train/images/
│   ├── train/labels/
│   ├── valid/images/
│   └── valid/labels/
├── roboflow_test/           # Hand-drawn wires (140 images)
│   ├── train/images/
│   ├── train/labels/
│   └── valid/images/
│   └── valid/labels/
├── Database/                # Raw schematic images (662 images)
│   ├── 001-099/
│   ├── 100-199/
│   └── ...
└── data/synthetic/          # Generated synthetic dataset
    ├── images/
    └── labels/

2. Generate Synthetic Dataset¶

uv run wire-sdg \
  --num-images 2000 \
  --image-size 1024 1024 \
  --output-dir data/synthetic

3. Configure Paths¶

Edit wire_detection/config/datasets.yaml to point to your dataset locations:

datasets:
  hdc:
    path: ./roboflow_test2          # relative to project root
    image_glob: "**/images/*.jpg"

  hand_drawn:
    path: ./roboflow_test
    image_glob: "**/images/*.jpg"

  synthetic:
    path: ./data/synthetic
    image_glob: "images/*.jpg"

  database:
    path: ./Database
    image_glob: "*/*.jpg"

4. Verify¶

# Check that datasets are found
curl http://localhost:8000/api/datasets

Expected response:

{
  "hdc": { "image_count": 1993, ... },
  "hand_drawn": { "image_count": 140, ... },
  "database": { "image_count": 662, ... }
}

Docker Setup¶

When using Docker Compose, datasets are mounted from the host into the container:

services:
  backend:
    volumes:
      - ./roboflow_test2:/data/hdc
      - ./roboflow_test:/data/hand_drawn
      - ./Database:/data/database
    environment:
      - DATASETS_YAML=/app/wire_detection/config/datasets.docker.yaml

The Docker config (datasets.docker.yaml) uses /data/<key> paths that match these mount points. No additional configuration needed.

Dataset Directory Layout¶

hdc/                          # roboflow_test2/
├── train/
│   ├── images/               # *.jpg
│   └── labels/               # *.txt  (YOLO OBB: class_id x1 y1 x2 y2 x3 y3 x4 y4)
├── valid/
│   ├── images/
│   └── labels/
└── test/
    ├── images/
    └── labels/

hand_drawn/                   # roboflow_test/
├── train/
│   ├── images/
│   └── labels/
└── valid/
    ├── images/
    └── labels/

database/                     # Database/
├── 001-099/                  # Raw circuit images
├── 100-199/
└── ...

synthetic/                    # data/synthetic/
├── images/                   # Generated images
└── labels/                   # Generated labels (lines format)

Environment Configuration¶

Use the DATASETS_YAML environment variable to select a different config file:

# Local development (default)
export DATASETS_YAML=wire_detection/config/datasets.yaml

# Docker (handled automatically by docker-compose.yml)
export DATASETS_YAML=/app/wire_detection/config/datasets.docker.yaml