Image Classification Model Training via Dynamic Recall-Based Weight Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer vision tasks on edge devices face challenges such as bandwidth limitations, latency issues, and data set imbalance, particularly in image classification, where standard loss functions tend to bias towards large classes, resulting in low accuracy and precision for small classes.
Innovation Solution
A method for training an image classification model that dynamically adjusts the weights of small classes by using recall rates from previous training iterations to balance accuracy across classes, reducing computational cost and addressing data set imbalance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard loss functions are used for image classification training, then training process is simple, but accuracy for small classes deteriorates due to dataset imbalance
Solution Approach 1:
The patent implements dynamic weight adjustment for the loss function based on recall rates from previous training iterations. The weight for each class is updated iteratively using the formula: weight_t(class) = weight_{t-1}(class) × recall_rate_{t-1}(class), where t represents the current iteration. This dynamic adjustment allows the training process to adaptively balance the weights of different classes, improving accuracy for small classes while maintaining a relatively simple training framework.
2Speed
If edge devices process computer vision tasks locally, then service speed improves, but computational cost increases due to limited device resources
Solution Approach 1:
The patent changes the parameter distribution in the training process by dynamically adjusting class weights based on recall rates. This parameter change optimizes the model to focus more on underrepresented small classes, improving their classification accuracy without requiring additional computational resources. The method achieves better performance by redistributing the training focus rather than increasing overall computational power, making it suitable for edge devices with limited resources.
3Measurement precision
If training focuses on improving small class accuracy, then classification precision for small classes improves, but training time increases due to iterative weight adjustment
Solution Approach 1:
The patent implements a feedback mechanism where the recall rate from each training iteration is used to adjust the loss function weights for the next iteration. The feedback loop continuously monitors classification performance on small classes and adjusts training priorities accordingly. This feedback-driven approach efficiently directs computational resources toward improving small class accuracy without requiring excessive training time, as the system automatically adapts based on performance metrics.
Data Source
AI summary
Embodiments disclosed herein relate to a method, a device, and a computer program product for training an image classification model. The method includes: determining a first classification result obtained by the image classification model on pixel points of a sample image in a first training iteration, wherein the first classification result indicates that each of the pixel points belongs to one of a plurality of object classes; determining a second classification result obtained by the image classification model on the pixel points of the sample image in a second training iteration; determining, based on the first classification result and the second classification result, a first set of recall rates and a second set of recall rates for the plurality of object classes; and adjusting, based on the first set of recall rates and the second set of recall rates, the image classification model to obtain a trained image classification model.