Data Augmentation via Point Cloud Object Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Insufficient training data can hinder the convergence speed and performance of neural network-based models, necessitating data augmentation techniques to generate new data.
Innovation Solution
A method that augments training data by naturally synthesizing images of objects from a point cloud, using region extraction, target object determination, and image synthesis based on location information and point cloud data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed using traditional techniques (rotation, color change), then the training data quantity increases, but the data quality and naturalness deteriorate
Solution Approach 1:
The patent creates synthetic training data by copying and combining elements from existing real-world point cloud data and images. Instead of artificially transforming data, the system extracts objects from point clouds, generates realistic images of these objects, and composites them into synthetic training samples that maintain natural appearance and physical consistency.
Solution Approach 2:
The patent introduces an intermediary object bank that stores extracted objects from point cloud data. This intermediary repository enables the system to selectively combine real object data with synthetic backgrounds, creating augmented training data that maintains both realism and diversity without the artifacts of traditional transformation methods.
2Reliability
If more training data is generated through synthesis, then model performance improves, but computational complexity increases
Solution Approach 1:
The patent performs preliminary extraction and storage of objects from point cloud data into an object bank during an offline preprocessing stage. This preliminary action reduces the computational burden during training, as the system only needs to retrieve and composite pre-extracted objects rather than performing complex extraction and processing during each training iteration.
Solution Approach 2:
The patent segments point cloud data into individual objects and stores them separately in an object bank. This segmentation enables efficient retrieval and combination of specific objects with different backgrounds, reducing computational complexity by avoiding the need to process entire point cloud scenes during training data generation.
3Productivity
If synthetic data is generated using simple transformation methods, then processing speed increases, but data realism decreases
Solution Approach 1:
The patent copies actual objects from real point cloud data and uses their authentic geometric and textural properties to generate synthetic images. This copying approach maintains realism by preserving the actual appearance characteristics of objects rather than relying on simplified transformation models that compromise visual fidelity.
Data Source
AI summary
A method and apparatus with data augmentation are disclosed. The a method includes: based on information about objects included in target data, extracting a region for object synthesis from a point cloud of the target data; determining a target object based on location information about the extracted region; based on a point cloud of the target object and the point cloud of the target data, synthesizing the point cloud of the target object with the extracted region to generate a synthetic point cloud; and generating a synthetic image by synthesizing an image of the target object with an image of the target data based on the location information about the extracted region and the point cloud of the target object, wherein the synthetic point cloud and the synthetic image form an augmented training item.