Training data splitting method and electronic device
AssigneeAcer Incorporated
InventorChun-Lin Han
Life of this patent
Estimating the term…
Abstract
The disclosure provides a training data splitting method and an electronic device. The method includes the following steps. Training data including a plurality of training images is obtained. A plurality of feature vectors of the training images are obtained by using an encoder of a pre-trained model. Feature similarity between any two of the feature vectors is evaluated. A plurality of target training images are selected from the training images based on feature similarities associated with each feature vector. The training data is split into a training set including the target training image and a verification set, wherein the training set and the verification set are utilized to train a machine learning model.
Claims18
A training data splitting method, comprising:
- obtaining training data comprising a plurality of training images;
- obtaining a plurality of feature vectors of the training images by using an encoder of a pre-trained model as a feature extractor, wherein the pre-trained model comprises an autoencoder;
- evaluating a feature similarity between any two of the feature vectors;
- selecting a plurality of target training images from the training images according to the feature similarities associated with each of the of feature vectors;
- and
- splitting the training data into a training set comprising the target training images and a validation set, wherein the training set and the validation set are utilized to train a machine learning model,
- wherein the step of selecting the target training images from the training images according to the feature similarities associated with each of the feature vectors comprises:
- calculating a similarity sum of the feature similarities between each feature vector and the other feature vectors;
- identifying a plurality of outlier feature vectors from the feature vectors based on the similarity sum corresponding to each of the feature vectors;
- and
- selecting the target training images corresponding to the outlier feature vectors from the training images.
An electronic device, comprising:
- a storage device, configured to store program code;
- and
- a processor, coupled to the storage device and configured to accesses the code to:
- obtain training data comprising a plurality of training images;
- obtain a plurality of feature vectors of the training images by using an encoder of a pre-trained model as a feature extractor, wherein the pre-trained model comprises an autoencoder;
- evaluate a feature similarity between any two of the feature vectors;
- select a plurality of target training images from the training images according to the feature similarities associated with each of the of feature vectors;
- and
- split the training data into a training set comprising the target training images and a validation set, wherein the training set and the validation set are utilized to train a machine learning model,
- wherein the processor is further configured to:
- calculate a similarity sum of the feature similarities between each feature vector and the other feature vectors;
- identify a plurality of outlier feature vectors from the feature vectors based on the similarity sum corresponding to each of the feature vectors;
- and
- select the target training images corresponding to the outlier feature vectors from the training images.
Dependencies are inferred from the claim text.