/aspera_dataset_generation

Primary LanguagePython

📝 Table of Contents

🧐 About

This repository allows you to generate synthetic images, annotated spv2 dataset format into spv2-COCO format, and enhance synthetic images using cycleGAN turbo.

Components

This repository contains :

  1. Syntethic_generation\: Utilizes Blender to generate synthetic images with their pose in SPV2 format.
  2. spv2_2_coco_annotation\: Converts and annotates SPV2 format dataset into the COCO format for keypoint detection (pose estimation).
  3. domain_adaptation\: Provide tools to translate images from synthetic domain to "lightbox" or "sunlamp" domain. (cycleGAN-turbo)
  4. data\: Contains both source and created datasets.

Everything can be runned locally exepted the training and inferencing of the AI-based model (cycleGAN-turbo here). They are performed using AWS (SageMaker and S3), dedicated scripts are provided.

🏁 Getting Started

These instructions will guide you through setting up your local environment to work with the project.

Prerequisites

Ensure you have the following software installed:

  • Git: For cloning and managing the repository.
  • Blender: Required for generating synthetic images and annotating datasets.
  • Python: The programming language used in the project, along with pip for package management.
  • Virtual Environment: To manage Python dependencies.
  • Git LFS: To handle large files in the repository.

Installing

Cloning the Repository

  1. Open your terminal.

  2. Navigate to the directory where you want to clone the project.

  3. Clone the repository using Git:

    git clone https://gitlab.sudouest.sii.fr/community/tmd/co-operate/aspera/aspera_synthetic_generation.git

  4. Navigate into the project directory:

    cd aspera_synthetic_generation

Blender

Blender is needed for dataset generation. To install Blender:

  1. Visit the official Blender download page.

  2. Download and install the appropriate version for your operating system.

  3. Confirm the installation by running:

    blender --version

Python, Pip, and Virtual Environment

Set up Python and its dependencies as follows:

  1. Ensure Python is installed. Download it from the official Python website.

  2. Verify pip installation:

    pip --version

  3. Create a virtual environment in your project directory:

    python -m venv venv

  4. Activate the virtual environment:

    • Windows:

      venv\Scripts\activate

    • macOS/Linux:

      source venv/bin/activate

  5. Install project dependencies from the requirements.txt file:

    pip install -r requirements.txt

Git Large File Storage (Git LFS)

Install and configure Git LFS for managing large files:

  1. Follow the Git LFS installation guide.

  2. Initialize Git LFS in your cloned repository:

    git lfs install

  3. Pull the LFS files to your local repository:

    git lfs pull

🎈 Usage

Dataset Generation

To generate a 3D dataset using Blender, navigate to the synthetic_generation directory and refer to the spv2_annotation/README.md file located there.

SPV2 2 COCO Annotation

To convert and annotate the SPV2 dataset to the COCO format, navigate to the spv2_2_coco_annotation directory and refer to the spv2_2_coco_annotation/README.md for detailed steps and usage.

Domain Adaptation

To convert synthetic images into "lightbox" or "sunlamp" domain images, navigate to the spv2_annotation directory and refer to the domain_adaptation/README.md for detailed steps and usage.

📦 Documentation