Synthetic Configuration

Explanation of Individual Pipeline Tasks

Create Recipes

Purpose:
Generates synthetic image recipes by selecting plant cutouts based on filtering criteria from MongoDB and associating them with background images. These recipes specify how synthetic images will be created.

Key Features:
- MongoDB Integration: Retrieves cutouts using specific filters from the MongoDB collection.
- Randomized Synthetic Images: Associates cutouts with randomly selected background images for diverse datasets.
- Cutout Usage Control: Configurable to reuse cutouts across multiple synthetic images or ensure unique usage.
- JSON Output: Saves generated recipes in JSON format for further processing.
Output:
- JSON files containing the list of synthetic image recipes, saved in the <repo_root>/projects/<project_name>/<sub_project_name>/recipes directory.

Move Cutouts

Purpose:
This task downloads plant cutout images from S3 bucket to a local directory, enabling the use of these images for generating synthetic data.

Output:
- Cutout images in .png format are stored locally for further use. Stored in <repo_root>/data/cutouts

Synthesize

Purpose:
Generates synthetic images by placing plant cutouts onto background images with to simulate real-world variability.

Key Features:
- Parallelism: Uses concurrent.futures.ProcessPoolExecutor to generate images in parallel.
- Transformations: Applies rotations, flips, and other transformations via Albumentations to enhance diversity.
- YOLO Annotations: Generates bounding box or contour annotations in YOLO format.
Configuration Options:
- Resize Factor: Control the scaling of cutouts (resize_factor: 0.35).
- Parallel Processing: Enable parallel synthesis (parallel: true or false).
- Mask Generation: Control the output of instance masks and YOLO labels.
Output:
- Images: Synthetic images saved as .jpg.
- YOLO Labels: Bounding box or contour labels for object detection.

Main Configuration Explanation Table

Parameter	Description	Possible Values / Example

Parameter	Description	Possible Values / Example
general.project_name	The name of the main project.	String (e.g., `my_test`)
general.sub_project_name	The name of the sub-project or task.	String (e.g., `my_sub_test`)
tasks	List of individual tasks to execute as part of the pipeline.	Example: `create_recipes`, `move_cutouts`, `synthesize`
synthesize	Configuration settings specific to the synthesize task.
synthesize.resize_factor	Resize factor for cutouts; controls cutout scaling during synthesis.	Example: `0.35` (Values < 0.15 may cause issues with `RandomScale` transformation)
synthesize.parallel	Enables or disables parallel processing for synthesis.	`true` or `false`
synthesize.parallel_workers	Number of workers to use for parallel processing.	Integer (e.g., `4`)
yolo_bbox_labels	Whether to generate YOLO-format bounding box labels.	`true` or `false`
aws.s3_bucket	The name of the AWS S3 bucket where data is stored.	Example: `psi-hackathon`
mongodb.host	Host address of the MongoDB database.	Example: `40.76.253.124`
mongodb.port	Port number for MongoDB.	Example: `27017`
mongodb.db	MongoDB database name.	Example: `hackathon_db`
mongodb.collection	MongoDB collection name for cutouts.	Example: `cutouts`
mongodb.auth_mechanism	MongoDB authentication mechanism and encryption method	Example: `SCRAM-SHA-1`
mongodb.auth_source	MongoDB database that the collection with user credentials	Example: `hackathon_db`
mongodb.username	MongoDB username	Example: `test_user`
mongodb.password	MongoDB password	Example: `pass123`

Cutout Filters Configuration Table

Parameter	Description	Possible Values / Example

Parameter	Description	Possible Values / Example
total_images	Total number of synthetic images to generate.	Integer (e.g., `1000`)
cuts_n_image	Number of plant cutouts to include in each image.	min: Minimum cutouts (e.g., `50`) max: Maximum cutouts (e.g., `200`)
reuse_cutouts	Whether the same cutouts can be reused in multiple images.	`true` or `false`
morphological.area	Range of the cutout area in pixels.	min: `10`, max: `10000000000`
morphological.blur_effect	Threshold range for cutout blur effect.	min: optional, max: optional
morphological.eccentricity	Shape eccentricity range.	min: optional, max: optional
morphological.extends_border	Whether to include only cutouts that extend the image border.	`true` or `false`
morphological.num_components	Range for the number of components in a cutout.	min: optional, max: optional
morphological.solidity	Threshold for solidity, indicating shape compactness.	min: `0.7` (example), max: optional
morphological.green_sum	Range for the sum of green pixel values in the cutout.	min: optional, max: optional
morphological.is_primary	Whether to include only primary cutouts.	`true` or `false` (if used)
morphological.perimeter	Range for cutout perimeter length.	min: optional, max: optional
category.family	Exact family name to filter by.	String (e.g., `Poaceae`)
category.genus	Exact genus name to filter by.	String (e.g., `Setaria`)
category.group	Plant group to filter by.	String (e.g., `Cover crop`)
category.duration	Plant lifecycle duration to include.	List (e.g., `["annual", "biennial"]`)
category.growth_habit	Filter based on growth habit of plants.	List (e.g., `["graminoid", "forb/herb"]`)
category.species	Exact species name to filter by.	String (e.g., `Setaria faberi`)
category.subclass	Exact subclass to filter by.	String (e.g., `Magnoliidae`)
category.common_name	List of plant common names to include.	Example: `["Black oats", "Goosegrass", "Maize"]`
common_name_weights	Assign weights to plant common names for frequency control.	Example: `Maize: 0.1, Goosegrass: 0.1`

Scripts/Tasks and How to Run Them in the SemiF-SyntheticPipeline

The SemiF-SyntheticPipeline offers flexible ways to run tasks by either adjusting the configuration files or overriding specific parameters from the command line. Below is an explanation of the tasks, how they work, and examples of how to run them.

How to Run the Pipeline

Using Configuration Files:
Adjust the configuration settings in the provided YAML files, then run the entire pipeline using the following command:
python main.py
Directly from the Command Line:
Override individual parameters while running specific tasks. This allows quick modifications without editing the configuration files.
Example:
python main.py synthesize.resize_factor=0.5 synthesize.parallel=true

Example Workflows

Full Pipeline Execution:
Adjust settings in the YAML configuration files and run the entire pipeline:
python main.py
Run Only the Synthesize Task with Custom Parameters:
If you need to modify specific settings (e.g., resize factor) without editing the config file:
Generate Recipes Only:
If you only need to create recipes:
Move Cutouts and Synthesize Images Sequentially:
Run move_cutouts and synthesize tasks in sequence with specific parameters:

Conclusion

The SemiF-SyntheticPipeline offers participants flexibility to adjust configurations through YAML files or directly from the command line, making it easy to experiment with different setups during the hackathon. This modular design ensures you can focus on specific tasks like generating recipes, moving cutouts, or synthesizing images, while optimizing performance using parallel processing and seamless AWS and MongoDB integrations.

Synthetic image

Hackathon-2024