View source for 17978425. METHOD AND APPARATUS WITH RECOGNITION MODEL TRAINING simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

=METHOD AND APPARATUS WITH RECOGNITION MODEL TRAINING=


==Organization Name==

[[:Category:SAMSUNG ELECTRONICS CO., LTD.|SAMSUNG ELECTRONICS CO., LTD.]]


[[Category:SAMSUNG ELECTRONICS CO., LTD.]]

==Inventor(s)==

[[:Category:Huijin Lee of Pohang-si (KR)|Huijin Lee of Pohang-si (KR)]][[Category:Huijin Lee of Pohang-si (KR)]]

[[:Category:Wissam Baddar of Suwon-si (KR)|Wissam Baddar of Suwon-si (KR)]][[Category:Wissam Baddar of Suwon-si (KR)]]

[[:Category:Minsu Ko of Suwon-si (KR)|Minsu Ko of Suwon-si (KR)]][[Category:Minsu Ko of Suwon-si (KR)]]

[[:Category:Sungjoo Suh of Seongnam-si (KR)|Sungjoo Suh of Seongnam-si (KR)]][[Category:Sungjoo Suh of Seongnam-si (KR)]]

==METHOD AND APPARATUS WITH RECOGNITION MODEL TRAINING - A simplified explanation of the abstract==


This abstract first appeared for US patent application 17978425 titled 'METHOD AND APPARATUS WITH RECOGNITION MODEL TRAINING

==Simplified Explanation==

The patent application describes a method for training an encoding model and a relationship estimation model using data augmentation and feature extraction techniques. The method involves generating sample images, extracting feature maps, determining loss data, estimating geometric information, and training the models based on the loss data.

* The method involves generating sample images by applying data augmentation techniques to an input training image.
* Feature extraction is performed on the sample images using an encoding model to generate feature maps.
* Loss data is determined based on the relationship between the feature vectors of the sample images.
* Relative geometric information of the feature maps is estimated using a relationship estimation model.
* Second loss data is determined based on the relative geometric information and label data.
* The encoding model and the relationship estimation model are trained using the first and second loss data.

== Potential Applications ==
* This method can be applied in various computer vision tasks such as object recognition, image classification, and image segmentation.
* It can be used in training models for autonomous vehicles to improve their perception and understanding of the environment.
* The method can be utilized in medical imaging to enhance the accuracy of diagnosis and analysis.

== Problems Solved ==
* The method addresses the problem of limited training data by generating additional sample images through data augmentation.
* It solves the problem of extracting meaningful features from the sample images by using an encoding model.
* The method also solves the problem of estimating the relative geometric information between the feature maps to improve the training process.

== Benefits ==
* By generating additional sample images, the method increases the diversity and quantity of training data, leading to improved model performance.
* The use of feature extraction and relationship estimation models enhances the ability to extract meaningful information and understand the geometric relationships between objects in the images.
* The training process of the encoding model and the relationship estimation model is optimized by considering both the feature vectors and the geometric arrangement of the sample images.


==Original Abstract Submitted==

A processor-implemented method includes: generating a first sample image and a second sample image by performing data augmentation on an input training image; generating a first feature map of the first sample image and a second feature map of the second sample image by performing feature extraction on the first sample image and the second sample image using an encoding model; determining first loss data according to a relationship between first feature vectors of the first feature map and second feature vectors of the second feature map; estimating relative geometric information of the first feature map and the second feature map using a relationship estimation model; determining second loss data according to the relative geometric information, based on label data according to a geometric arrangement of the first sample image and the second sample image in the input training image; and training the encoding model and the relationship estimation model, based on the first loss data and the second loss data.



[[Category:G06V10/774]]
[[Category:G06V10/77]]
[[Category:G06V10/82]]