System and method for generating training data for computer vision systems based on image segmentation
A system and method for training a model using a training dataset. The training dataset can be made up of only real data, only synthetic data, or any combination of synthetic data and real data. The images are segmented to define objects with known labels. The object is pasted onto backgrounds to generated synthetic datasets. The various aspects of the invention include generation of data that is used to supplement or augment real data. Labels or attributes can be automatically added to the data as it is generated. The data can be generated using seed data. The data can be generated using synthetic data. The data can be generated from any source, including the user's thoughts or memory. Using the training dataset, various domain adaptation models can be trained.
1. A method for segmenting images to generate a training dataset, the method comprising:
collecting a plurality of images;
segmenting each image of the plurality of images to identify a plurality of objects within each image to produce a plurality of segmented images, wherein at least one object of the plurality of objects is a target for training and the at least one object includes a known label;
pasting the at least one object onto different backgrounds; and
generating synthetic dataset to use as a training dataset to train a model for recognition of the target.
2. The method of claim 1 , further comprising altering the plurality of segmented images based on least one parameter.
3. The method of claim 1 , wherein the plurality of images are real.
4. The method of claim 1 , wherein the plurality of images are synthetic.
5. The method of claim 1 further comprising refining the synthetic dataset.
6. The method of claim 1 further comprising altering at least one segmented image selected from the plurality of labeled segmented images.
7. A non-transitory computer readable medium for storing code that, when said code is executed by a system, would cause the system to:
collecting a plurality of images;
segmenting each image of the plurality of images to identify a plurality of objects within each image to produce a plurality of segmented images, wherein at least one object of the plurality of objects is a target for training and the at least one object includes a known label;
pasting the at least one object onto different backgrounds; and
generating synthetic dataset to use as a training dataset to train a model for recognition of the target.
8. The non-transitory computer readable medium of claim 7 for storing code that, when executed, would further cause the system to alter the segmented image by altering at least one parameter.
9. The non-transitory computer readable medium of claim 7 , wherein the images are real images.
10. The non-transitory computer readable medium of claim 7 , wherein the images are synthetic images.
11. A method for neural network training, the method comprising:
capturing real data, wherein the real data includes a plurality of real images;
segmenting each of the plurality real images;
identifying a plurality of objects in the segmented images;
generating a plurality of synthetic images using the plurality of objects;
labeling each object in the plurality of synthetic images;
augmenting the plurality of real images with the plurality of synthetic images to generate a training dataset; and
training the neural network using the training dataset.
12. The method of claim 11 , wherein the step of generating includes pasting an object onto a background.
13. The method of claim 12 , wherein the step of generating includes using a generative machine learning model.
14. The method of claim 13 , wherein the generative machine learning model uses a generative adversarial network (GAN).
15. The method of claim 13 , wherein the generative machine learning model uses at least one generative model including a computer generated imagery (CGI), a variational auto encoder (VAE), and a pixel-by-pixel generative model.
16. The method of claim 11 , wherein the segmented images include known labels.
17. The method of claim 11 , wherein the plurality of synthetic images include parameters that mimic parameters of the plurality of real images.