US 11,893,774 B2
Systems and methods for training machine models with augmented data
Matthew John Cooper, Providence, RI (US); Paras Jagdish Jain, Cupertino, CA (US); and Harsimran Singh Sidhu, Fremont, CA (US)
Assigned to Tesla, Inc., Austin, TX (US)
Filed by Tesla, Inc., Austin, TX (US)
Filed on Dec. 14, 2021, as Appl. No. 17/644,308.
Application 17/644,308 is a continuation of application No. 16/598,956, filed on Oct. 10, 2019, granted, now 11,205,093.
Claims priority of provisional application 62/744,534, filed on Oct. 11, 2018.
Prior Publication US 2022/0108130 A1, Apr. 7, 2022
Int. Cl. G06V 10/772 (2022.01); G06F 18/214 (2023.01); G06F 18/213 (2023.01); G06V 10/774 (2022.01); G06V 20/00 (2022.01); G06V 20/56 (2022.01)
CPC G06V 10/772 (2022.01) [G06F 18/213 (2023.01); G06F 18/214 (2023.01); G06F 18/2148 (2023.01); G06V 10/774 (2022.01); G06V 20/00 (2022.01); G06V 20/56 (2022.01)] 17 Claims
OG exemplary drawing
 
1. A method for training a set of parameters of a predictive computer model, the method comprising:
obtaining a plurality of images and associated training outputs, the images being captured by cameras of one or more vehicles;
for an individual image of the plurality of images, generating an augmented image for the individual image based on modifying the individual image with an image manipulation function of one or more image manipulation functions which maintain camera properties of the individual image, such that angle, scale, and/or pose associated with the individual image is preserved, wherein the augmented training image is associated with the associated training output of the individual image, and
wherein the one or more image manipulation functions include a cutout function which adjusts a portion of an image based on a region of interest, wherein the image is obtained via a particular camera, and wherein the region of interest corresponds to artifacts which are always present in images obtained via the particular camera; and
training the predictive computer model based, at least, on the individual image and the augmented image, wherein the trained predictive computer model is configured to predict a presence of objects in input images for use in autonomous or semi-autonomous control of a particular vehicle.