Architecture¶

The framework is organized as two top-level packages and supporting infrastructure:

LineDetection/
├── wire_detection/          # Python backend (pipeline, sdg, evaluate, experiment, api)
│   ├── pipeline/            # Composable detection pipeline
│   ├── api/                 # FastAPI server
│   ├── data/                # Dataset registry with YAML config
│   ├── sdg/                 # Synthetic data generator
│   ├── evaluate/            # Evaluation metrics and reporting
│   ├── experiment/          # Parameter sweep engine
│   ├── config/              # Dataset YAML configs
│   └── tests/               # Test suite
├── ui/                      # Next.js frontend
├── docker-compose.yml       # Orchestrates both services
├── Dockerfile.python        # Python backend container
└── pyproject.toml           # Package metadata and CLI entry points

Backend (`wire_detection/`)¶

Pipeline Module¶

The core detection pipeline — a chain of independent stages. Each stage implements the PipelineStage ABC:

class PipelineStage(ABC):
    name: str
    def run(self, image: np.ndarray, params: dict) -> StageOutput: ...
    def visualize(self, image: np.ndarray, output: StageOutput) -> np.ndarray: ...

Stages are composed by the Pipeline class and built from config via PipelineFactory. See Pipeline Overview.

API Module¶

FastAPI server with CORS, LRU image cache, and endpoints for listing images, serving thumbnails, running the pipeline, and querying available stages/datasets. See API Endpoints.

Data Module¶

Dataset registry that resolves paths, validates structure, and normalizes labels. Supports multiple label formats (YOLO OBB, YOLOv8 pose). See Configuration.

SDG Module¶

Generates realistic circuit schematic images with bezier-curve wires, component boxes, paper textures, and tool strokes. Exports labels in multiple formats. See SDG.

Evaluate Module¶

Line-distance metric with greedy matching for comparing detected lines against ground truth. Generates per-image and aggregate reports. See Evaluation.

Experiment Module¶

Grid and random parameter search engine with checkpointing and markdown ranking tables. See Experiment Engine.

Frontend (`ui/`)¶

Next.js 14 app with TypeScript, shadcn/ui components, and dark theme. Features:

Sidebar — dataset selector, image picker, all pipeline parameter sliders, Run button, stats
4-panel grid — Detected Lines, Threshold, Dilated, Source — click for fullscreen
Image picker — full-screen modal with 5-column thumbnail grid, lazy-loaded
Live updates — slider release triggers pipeline run with abortable requests

CLI Entry Points¶

Defined in pyproject.toml and installed as console scripts:

Command	Module	Description
`wire-tune`	`api.server:main`	Start the FastAPI tuner server
`wire-pipeline`	`pipeline.cli:main`	Run pipeline on a single image
`wire-sdg`	`sdg.cli:main`	Generate synthetic dataset
`wire-eval`	`evaluate.cli:main`	Evaluate detections against ground truth
`wire-sweep`	`experiment.cli:main`	Run a parameter sweep

Data Flow¶

Config (YAML/dict) → PipelineFactory → Pipeline (composed stages)
                                         ↓
Image → [crop → mask → threshold → invert → dilate → CCL → contour_extract → dedup → length_filter]
                                         ↓
                                  PipelineResult (lines, intermediate images, timing)
                                         ↓
                              Optional: Evaluate against GT → EvalResult (F1, precision, recall)