Point Cloud Neural Network Accelerator with Selective Distance Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud processing techniques, such as PointNet and PointNet++, require significant computational resources for sampling and grouping operations due to the randomness of point cloud data, leading to inefficient calculation times, especially when dealing with large datasets.
Innovation Solution
An accelerator for point cloud deep neural networks that includes a distance calculator with serial and parallel operation modes to calculate distances on three axes, allowing for reduced calculations and energy consumption by omitting unnecessary distance computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all three types of 3D information are used for distance calculation, then measurement precision is improved, but calculation time increases
Solution Approach 1:
The patent applies partial action by selectively using only one type of 3D information (Euclidean distance, Chebyshev distance, or Manhattan distance) instead of all three types for distance calculation. This partial usage reduces the computational burden while maintaining sufficient precision for the sampling and grouping operations in point cloud processing.
2Measurement precision
If distance calculations are performed for all point pairs, then measurement precision is improved, but energy consumption increases
Solution Approach 1:
The patent extracts and uses only the necessary distance information required for sampling and grouping operations. By taking out only the essential distance calculations needed for the task rather than computing all possible distance pairs, the energy consumption is significantly reduced while maintaining the required measurement precision.
3Manufacturing precision
If sampling and grouping processes are performed with high precision, then processing accuracy is improved, but productivity decreases
Solution Approach 1:
The patent changes the parameter of distance calculation by providing multiple distance metric options (Euclidean, Chebyshev, Manhattan) and allowing selection based on the specific application requirements. This parameter change enables balancing between processing accuracy and calculation speed, improving productivity while maintaining sufficient precision for the task.
Data Source
AI summary
An accelerator for processing a point cloud deep neural network according to an aspect of the present specification includes at least one distance calculator configured to calculate a distance between two points different from each other, wherein the distance calculator includes a first calculation unit, a second calculation unit, and a third calculation unit, and the first calculation unit, the second calculation unit, and the third calculation unit are driven serially to calculate distances on three axes between the two points in a first operation mode, and are driven in parallel to respectively calculate the distances between different pairs of points, in a second operation mode.