SemiF-SyntheticPipeline

Installation and Setup

Installing Conda

To manage the project's dependencies efficiently, we use Conda, a powerful package manager and environment manager. Follow these steps to install Conda if you haven't already:

1. Download the appropriate version of Miniconda for your operating system from the official Miniconda website.
2. Follow the installation instructions provided on the website for your OS. This typically involves running the installer from the command line and following the on-screen prompts.
3. Once installed, open a new terminal window and type conda list to ensure Conda was installed correctly. You should see a list of installed packages.

Setting Up Your Environment Using an Environment File

After installing Conda, you can set up an environment for this project using an environment file, which specifies all necessary dependencies. Here's how:

1. Clone this repository to your local machine.

2. Navigate to the repository directory in your terminal.

3. Locate the environment.yaml file in the repository. This file contains the list of packages needed for the project.

4. Create a new Conda environment by running the following command:

   conda env create -f environment.yaml

   This command reads the environment.yaml file and creates an environment with the name and dependencies specified within it.

5. Once the environment is created, activate it with:

   conda activate <env_name>

   Replace <env_name> with the name of the environment specified in the environment.yaml file.

Scripts:

Json to Mongo

This script loads JSON data from batch directories in an NFS storage system into a MongoDB database. It reads the batch names from a YAML configuration file, checks both primary and secondary NFS storage locations for the corresponding JSON metadata files, and inserts the data into a specified MongoDB collection.

Key Features

• MongoDB Integration: Connects to MongoDB to insert JSON data.
• Batch Processing: Reads batch names from a YAML configuration and processes the corresponding directories in the NFS storage locker.
• Primary and Secondary Storage: Automatically checks both primary and secondary NFS storage paths for the presence of batch directories.

Output

• Data Insertion: Inserts JSON data from batch directories into the specified MongoDB collection.

Create Recipes

This script is responsible for creating synthetic image recipes by selecting cutout images based on specific criteria and associating them with background images. The recipes are then saved in JSON format for use in synthetic dataset generation.

Key Features

• MongoDB Integration: Retrieves cutout metadata from a MongoDB collection based on specific filter criteria defined in the configuration.
• Randomized Synthetic Image Generation: Associates cutouts with randomly selected background images and creates synthetic images with varying numbers of cutouts.
• Flexible Cutout Usage: Configurable to either reuse cutouts across multiple synthetic images or ensure each cutout is used only once.
• JSON Output: Saves the generated synthetic image recipes to a JSON file for further processing.

Output

• Synthetic Image Recipes: A JSON file containing a list of synthetic images, each with a unique ID, background image, and associated cutouts. The file is saved in the recipes directory under the project directory.

Move Cutouts

This script is responsible for downloading plant cutout images from long-term storage to a local directory. It can handle both sequential and concurrent data transfer. The downloaded cutouts are stored locally for further use in synthetic image generation.

Key Features

• Sequential and Parallel Processing: The script can download cutouts in a sequential manner or use multithreading.
• Dual Storage Locations: Looks in both primary and secondary long-term storage locations.

Output

• Downloaded Images: The script downloads .png cutout images to the specified local directory.

Synthesize

This script is designed to generate synthetic images by overlaying plant cutout images onto various backgrounds using a copy-and-paste method. The script provides CPU parallelism.

• Parallelism: Utilizes Python's concurrent.futures.ProcessPoolExecutor to enable concurrent processing of multiple image recipes, leveraging multi-core CPUs.
• Transformations: Applies a variety of image transformations (e.g., rotation, flipping) using the Albumentations library.
• Dynamic Shadow Generation: Simulates dynamic shadows for the cutouts based on their size and position relative to the light source.
• Cutout Distribution: Supports random placement of cutouts on background images, creating diverse compositions.
• Output Flexibility: Saves images, semantic masks, instance masks, and YOLO format segmentation labels.

Output

• Images: Generated synthetic images in .jpg format.
• Semantic Masks: Corresponding masks with class annotations in .png format.
• Instance Masks: Optional masks for instance annotations.
• YOLO Labels: Segmentation contours in YOLO format.

License

This script is provided as-is, with no warranties or guarantees. You are free to modify and distribute it as needed. However, attribution is appreciated if you share it publicly.