Cycle GAN RGB to Non-RGB Image Transformation for Autonomous Driving

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for autonomous vehicles face challenges in processing test images in non-RGB formats due to the need for real-time format conversion, which is inefficient and requires extensive training data, especially when converting complex or non-existent formats, and re-learning CNN parameters.

Innovation Solution

A method using a cycle GAN to transform annotated RGB images into non-RGB images, incorporating a learning device with transformers and discriminators to achieve format conversion while optimizing the transformation process and reducing annotation costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If real-time format conversion is performed during testing processes, then the CNN can process non-RGB format images, but computational overhead increases significantly

Engineering Contradiction:
Improveformat compatibilityVSAvoidprocessing speed
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by performing format transformation during the training phase rather than during real-time testing. The cycle GAN is trained to transform RGB images to non-RGB formats and back, creating pre-learned transformation capabilities. This allows the CNN to directly process non-RGB format images during testing without performing computationally intensive real-time conversion, thus resolving the contradiction between format adaptability and processing speed.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If the CNN is re-trained with new training images in non-RGB format, then it can process those formats properly, but tremendous time and money are required to generate tens of thousands of training images and GTs

Engineering Contradiction:
Improveformat processing capabilityVSAvoidtraining time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent uses copying by generating synthetic non-RGB format training images through the cycle GAN transformation process. Instead of manually creating tens of thousands of annotated non-RGB images, the system copies and transforms existing RGB training images into non-RGB formats using the learned transformation model. This dramatically reduces the time and resources required for training data preparation while maintaining the ability to process non-RGB formats effectively.

Inventive Principle:
Principle #26Copying

3Productivity

If simple format conversion rules are applied (e.g., YUV to RGB), then conversion overhead is trivial, but complex conversion rules or non-existent rules cannot be applied

Engineering Contradiction:
Improveconversion efficiencyVSAvoidformat conversion capability
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent applies parameter changes by using the cycle GAN to learn and adapt transformation parameters for converting between different image formats. The GAN model adjusts its internal parameters during training to capture the specific characteristics of different format conversions, enabling efficient transformation even for complex or non-standard format rules that cannot be handled by simple predefined conversion algorithms.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3690737B1Learning method and learning device for strategic transforming RGB training image sets into non-RGB training image sets, to be used for learning object detection on objects of images in non-RGB format, by using cycle GAN, resulting in significantly reducing computational load and reusing data
Publication Date: 2022.08.17 STRADVISION
  • EP3690737B1 patent drawingFigure 1
  • EP3690737B1 patent drawingFigure 2
  • EP3690737B1 patent drawingFigure 3

AI summary

A method for learning transformation of an annotated RGB image into an annotated Non-RGB image, in target color space, by using a cycle GAN and for domain adaptation capable of reducing annotation cost and optimizing customer requirements is provided. The method includes steps of: a learning device transforming a first image in an RGB format to a second image in a non-RGB format, determining whether the second image has a primary or a secondary non-RGB format, and transforming the second image to a third image in the RGB format; transforming a fourth image in the non-RGB format to a fifth image in the RGB format, determining whether the fifth image has a primary RGB format or a secondary RGB format, and transforming the fifth image to a sixth image in the non-RGB format. Further, by the method, training data can be generated even with virtual driving environments.