Data augmentation for image classification tasks
View Patent ↗A computer-implemented method and systems are provided for performing machine learning for an image classification task. The method includes selecting, by a processor operatively coupled to one or more databases, a first and a second image from one or more training sets in the one or more databases. The method further includes overlaying, by the processor, the second image on the first image to form a mixed image, by averaging an intensity of each of a plurality of co-located pixel pairs in the first and the second image. The method also includes training, by the processor, a machine learning process configured for the image classification task using the mixed image to augment data used by the machine learning process for the image classification task.
1. A computer-implemented method for performing machine learning for an image classification task, the method comprising:
selecting, by a processor operatively coupled to one or more databases, a first and a second image from one or more training sets in the one or more databases;
overlaying, by the processor, the second image on the first image to form a mixed image, by averaging an intensity of each of a plurality of co-located pixel pairs in the first and the second image; and
training, by the processor, a machine learning process configured for the image classification task using the mixed image to augment data used by the machine learning process for the image classification task.
2. The computer-implemented method of claim 1 , wherein said training step comprises using a label of the first image as a label of the mixed image.
3. The computer-implemented method of claim 1 , wherein said training step comprises using a label of the second image as a label of the mixed image.
4. The computer-implemented method of claim 1 , wherein said training step comprises mixing a label of the first image and a label of the second image to form a label of the mixed image.
5. The computer-implemented method of claim 1 , wherein the selecting and overlaying steps are repeated for each of a plurality of different image pairs selected from the one or more training sets in the one or more databases to form a plurality of additional mixed images, and wherein the machine learning process is trained using the plurality of additional mixed images.
6. The computer-implemented method of claim 5 , further comprising performing additional training without using any further mixed images to increase a training speed of the machine learning process.
7. The computer-implemented method of claim 5 , wherein the additional training comprises a fine-tuning process for refining the machine learning process.
8. The computer-implemented method of claim 1 , wherein said selecting and overlaying steps are part of a data augmentation process used for training the machine learning process in said training step, and wherein said data augmentation process is selectively disabled and enabled at one or more time periods to increase a training speed.
9. The computer-implemented method of claim 8 , wherein the one or more time periods comprise multiple consecutive time periods, a first one of the multiple consecutive time periods corresponding to a commencement of a training stage for training the machine learning process.
10. The computer-implemented method of claim 8 , wherein the one or more time periods are at intermediate periods in a training stage for training the machine learning process.
11. The computer-implemented method of claim 1 , wherein the processor is a graphics processing unit.
12. The computer-implemented method of claim 1 , wherein the image classification task relates to an advanced driver-assistance system and the method further comprises:
applying the trained machine learning process to a test image to obtain a classification for the test image; and
controlling a function of one or more hardware devices of a motor vehicle, responsive to the classification for the test image.