Object Classifier Neural Network for Coarse and Fine Label Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately classifying objects using neural networks, as existing systems degrade in coarse-object classification performance when transitioning from coarse to fine-object classification tasks, leading to reduced accuracy and increased computational inefficiency.
Innovation Solution
The proposed solution involves training an object classification neural network in two phases, where the first phase optimizes coarse-object classifications and the second phase refines these classifications by freezing parameters of channel encoder subnetworks not essential for fine-object classifications, thereby maintaining accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the neural network is trained to perform fine-object classification, then the fine-object classification accuracy is improved, but the coarse-object classification performance degrades
Solution Approach 1:
The neural network is segmented into multiple independent channel encoder subnetworks, each specialized for specific object categories. This segmentation allows different parts of the network to learn different classification tasks independently, preventing the degradation of coarse-object classification when training for fine-object classification.
Solution Approach 2:
The network parameters are frozen before fine-object classification training to preserve the coarse-object classification capabilities that were learned in preliminary training phases. This preliminary freezing action prevents the degradation of earlier learned skills while new skills are acquired.
2Productivity
If all neural network parameters are adjusted during fine-object classification training, then the fine-object classification learning is accelerated, but the computational overhead increases
Solution Approach 1:
Different portions of the neural network have different training states - some parameters are frozen while others are actively trained. This local differentiation allows the network to focus computational resources only on the parameters necessary for fine-object classification, reducing overall computational overhead while maintaining learning effectiveness.
Solution Approach 2:
Instead of training all parameters, only the necessary parameters for fine-object classification are adjusted during training. This partial action approach reduces computational overhead while still achieving the desired learning objective for fine-grained classification.
Data Source
AI summary
Aspects of the subject matter disclosed herein include methods, systems, and other techniques for training, in a first phase, an object classifier neural network with a first set of training data, the first set of training data including a first plurality of training examples, each training example in the first set of training data being labeled with a coarse-object classification; and training, in a second phase after completion of the first phase, the object classifier neural network with a second set of training data, the second set of training data including a second plurality of training examples, each training example in the second set of training data being labeled with a fine-object classification.