multipark-aiml

Multipark AI/ML Platform

This repository contains tools and scripts for image segmentation, model training, inference, and dataset management using YOLO and Roboflow.

Features

• Download datasets from Roboflow interactively or via CLI
• Train YOLO segmentation models
• Run inference on images or folders (with GUI and CLI options)
• Deploy trained models to Roboflow
• Create and use polygon masks for region-of-interest inference
• Compare segmentation results with Aiforia outputs

Requirements

• Python 3.8+ (tested with 3.11)
• See image-segmentation/requirements.txt for Python dependencies

Setup

1. Clone the repository:

   git clone https://github.com/your-org/multipark-aiml.git
   cd multipark-aiml/image-segmentation

2. Install dependencies:

   pip install -r requirements.txt

3. Create a .env file in image-segmentation/ with your Roboflow API key and paths:

   ROBOFLOW_API_KEY="your_roboflow_api_key"
   DATASET_PATH="./datasets"
   MODEL_PATH="./models"

Usage

Download a Dataset

python download_dataset.py --api-key <API_KEY> [--project <project_slug>] [--version <version>] [--dataset-format yolov11] [--dataset-path <path>]

If --project is not provided, you will be prompted to select one interactively.

Train a YOLO Model

python train_yolo.py --dataset-dir <dataset_dir> [--epochs 100] [--imgsz 640] [--batch -1] [--device <device>] [--model yolo11n-seg]

Run Inference (CLI)

python infer_yolo.py --model <model_path> --input-path <image_or_folder> --output-path <output_folder> [options]

Key options:

• --no-colors: Convert output to grayscale
• --sliding-window: Use sliding window for large images
• --save-image: Save visualization
• --save-geojson: Save results as GeoJSON
• --use-geojson-mask: Use a polygon mask for ROI

Run Inference (GUI)

python gui_infer_yolo.py

A graphical interface will open for model selection and inference. (Tested on Mac, tested and not working on Linux CentOS.)

Deploy a Model to Roboflow

python deploy_model.py --model <model_path> --api-key <API_KEY> [--project <project_slug>] [--version <version>]

If --project is not provided, you will be prompted to select one interactively.

Create a Polygon Mask

python create_mask.py --image <image_path> [--output <output_geojson>]

Interactively draw a polygon on the image to create a GeoJSON mask.

Compare with Aiforia Results

python compare_aiforia.py --aiforia-csv <csv> --multipark-geojson <geojson> --output-path <dir> --pixel-size <size> [--image <img>] [--save-geojson] [--save-image]

Best Practices

• Keep your .env file in image-segmentation/ and add it to .gitignore.
• Do not commit sensitive information (API keys, etc.) to version control.
• Use a virtual environment for Python dependencies.

License

See LICENSE for details.

Contact

For questions or contributions, please open an issue or pull request.