Synthetic Training Image Generation for Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual annotation of images for training object recognition models is time-consuming, error-prone, and requires a large set of suitable images, which may not always be available, leading to biased and less accurate models, especially when recognizing objects in varying environments.
Innovation Solution
Using computer-generated training images that depict a digital training model of real-world objects, allowing for faster and more efficient training of object recognition models, and providing augmentations such as visual content, instructions, and virtual controls after recognition, enabling accurate object identification and manipulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual annotation of images is used to train object recognition models, then the models can be trained with real-world data, but the process is time-consuming and error-prone
Solution Approach 1:
The patent uses digital training models (virtual copies) of real-world objects to generate synthetic training images, replacing the need for manual annotation of physical objects. This copying approach maintains training effectiveness while eliminating time-consuming manual processes
Solution Approach 2:
The system performs preliminary actions by pre-creating digital training models and generating synthetic training images in advance. This prepares training data beforehand, eliminating the need for time-consuming manual annotation during the actual training process
2Adaptability or versatility
If a large set of annotated images is collected for training, then the model can be trained comprehensively, but the process becomes more complex and resource-intensive
Solution Approach 1:
The digital training models serve multiple functions: they can be rendered in various environments, lighting conditions, and perspectives to generate diverse training images from a single model. This universal approach replaces the need for collecting numerous specific annotated images
Solution Approach 2:
The system changes parameters such as lighting conditions, background environments, object positions, and camera angles when rendering synthetic training images from digital models. This generates diverse training data through parameter variation rather than collecting diverse physical samples
3Measurement precision
If manual annotation is performed by human annotators, then real-world objects can be accurately labeled, but human error and bias are introduced
Solution Approach 1:
The system uses automated rendering processes where digital training models generate their own training images with inherent labeling information. This self-service approach eliminates human annotators, removing the source of human error and bias while maintaining accuracy through precise digital modeling
Data Source
AI summary
A method for object recognition includes, at a computing device, receiving an image of a real-world object. An identity of the real-world object is recognized using an object recognition model trained on a plurality of computer-generated training images. A digital augmentation model corresponding to the real-world object is retrieved, the digital augmentation model including a set of augmentation-specific instructions. A pose of the digital augmentation model is aligned with a pose of the real-world object. An augmentation is provided, the augmentation associated with the real-world object and specified by the augmentation-specific instructions.