Image-to-DSL Model for Synthetic Plant Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of machine learning models for detecting and classifying plant traits in digital images is hindered by the high cost and scarcity of annotated training data, particularly for complex plant structures, as existing methods require manual annotation and extensive data collection.
Innovation Solution
An image-to-DL machine learning model is trained to translate ground truth images into parameterizable expressions in a domain-specific language like L-systems, enabling the automatic generation of synthetic training images that accurately represent real-world plant structures, reducing the need for manual annotation and increasing data availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation of training images is used, then training data accuracy is improved, but cost and time consumption increase significantly
Solution Approach 1:
The system performs preliminary actions by using a trained image-to-DSL model to automatically generate parameterized expressions from ground truth images before manual annotation is needed. These expressions serve as templates that can be rapidly instantiated to create synthetic training images with automatic annotations, eliminating the need for time-consuming manual pixel-wise annotation while maintaining accuracy through the structured parameterized representations.
Solution Approach 2:
The system creates copies by generating multiple synthetic training images from a single parameterized expression template. Once a ground truth image is converted to a DSL expression, this expression can be instantiated numerous times with different parameter values to produce varied synthetic images, all with automatically generated annotations, thereby multiplying the effective training data without additional manual annotation effort.
2Measurement precision
If more training data is collected to improve model accuracy, then detection precision is improved, but data acquisition cost and difficulty increase
Solution Approach 1:
The system addresses data scarcity by copying existing ground truth images through the image-to-DSL model to generate numerous synthetic variants. Each ground truth image serves as a seed that can produce multiple synthetic training images with different parameters while maintaining structural accuracy, effectively amplifying limited real-world data into large-scale training sets without requiring additional field data collection.
Solution Approach 2:
The system enables easy data multiplication by changing parameters within the DSL expressions. Once a ground truth image is converted to a parameterized expression, various parameters (such as plant size, orientation, lighting conditions, background elements) can be modified to generate diverse synthetic images, creating abundant training data with controlled variations that improve model robustness without requiring physical data collection.
3Quantity of substance
If DSL expressions are used to generate synthetic images, then data availability is improved, but ease of operation deteriorates due to difficulty in learning and writing DSL expressions
Solution Approach 1:
The system inverts the traditional workflow by training a machine learning model to translate images INTO DSL expressions, rather than requiring users to manually write DSL expressions from images. This reversal automates the previously manual step, allowing users to simply provide ground truth images and obtain parameterized DSL expressions automatically, eliminating the need for users to learn complex DSL syntax while still enabling synthetic image generation.
Solution Approach 2:
The system implements self-service by enabling automatic generation of DSL expressions from images through the trained model. The image-to-DSL model autonomously performs the translation task, selecting appropriate DSL commands and parameters without human intervention. This self-automating capability allows the system to generate training data independently, reducing manual effort to minimal operations such as providing input images and retrieving generated expressions.
Data Source
AI summary
Implementations are described herein for learning mappings between a domain specific language (DSL) and images, and leveraging those mappings for various purposes. In various implementations, a method for using a DSL to generate training data may include processing data indicative of ground truth image(s) depicting a real plant using a trained image-to-DSL machine learning (ML) model to generate a first expression in the DSL that describes structure of the real plant. The first expression may include a plurality of parameters, and may be processed to programmatically generate a plurality of synthetic DSL expressions. Each respective synthetic DSL expression may describe structure of a respective synthetic plant for which parameter(s) have been altered from the first expression. The synthetic DSL expressions may be processed using a renderer to create three-dimensional (3D) synthetic plant models. Two-dimensional (2D) synthetic images may be generated that depict the 3D synthetic plant models in an area.