Unsupervised Object Detection Training From Transformed Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning methods for object detection models require a large amount of annotated data, leading to high training costs.
Innovation Solution
An object detection model training method that utilizes unsupervised learning by transforming unannotated sample images and their prediction results, calculating a loss value based on the difference between transformed and predicted results, and iteratively adjusting model parameters until a training end condition is met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised learning method is used to train object detection model, then model training accuracy can be ensured, but training cost increases due to requirement of large amount of annotated data
Solution Approach 1:
The system performs self-annotation by generating pseudo-labels through its own prediction capabilities. The object detection model generates initial predictions on unannotated images, and these predictions are transformed and used as training labels, eliminating the need for manual annotation of training data while maintaining model accuracy through iterative self-improvement
Solution Approach 2:
The system performs preliminary prediction on unannotated images before formal training. By first generating prediction results on unannotated data and then using these predictions as basis for creating training samples, the system prepares training data in advance without requiring manual annotation, thus reducing training cost while ensuring subsequent training effectiveness
Data Source
AI summary
An object detection model training method includes: inputting an unannotated first sample image into an initial detection model of a current round, and outputting a first prediction result for a target object, transforming the first sample image and a first prediction position region within the first prediction result to obtain a second sample image and a prediction transformation result in the second sample image; inputting the second sample image into the initial detection model, and outputting a second prediction result for the target object; obtaining a loss value of unsupervised learning according to a difference between the second prediction result and the prediction transformation result; and adjusting model parameters of the initial detection model according to the loss value and returning to the operation of inputting a first sample image into an initial detection model of a current round to perform iterative training, to obtain an object detection model.