SYSTEM FOR TRAINING NETWORKS FOR SEMANTIC SEGMENTATION
First Claim
1. A device comprising:
- a processor; and
a computer-readable medium in communication with the processor, the computer-readable medium including one or more modules comprising;
a training supervisor module configured to initiate and control a neural network training process, the training supervisor module configured to receive a training image;
a mask generator module configured to;
generate candidate segment masks based on the training image received by the training supervisor module; and
rank the candidate segment masks to generate ranked candidate segment masks;
a mask selector module configured to select a set of the ranked candidate segment masks to generate a set of ranked candidate segment masks; and
a neural network updater module configured to;
select one of the set of ranked candidate segment masks to train the neural network; and
train the neural network.
1 Assignment
0 Petitions
Accused Products
Abstract
Disclosed herein are technologies directed to training a neural network to perform semantic segmentation. A system receives a training image, and using the training image, candidate masks are generated. The candidate masks are ranked and a set of the ranked candidate masks are selected for further processing. One of the set of the ranked candidate masks is selected to train the neural network. The one of the set of the set of the ranked candidate masks is also used as an input to train the neural network in a further training evolution. In some examples, the one of the set of the ranked candidate masks is selected randomly to reduce the likelihood of ending up in poor local optima that result in poor training inputs.
24 Citations
20 Claims
-
1. A device comprising:
-
a processor; and a computer-readable medium in communication with the processor, the computer-readable medium including one or more modules comprising; a training supervisor module configured to initiate and control a neural network training process, the training supervisor module configured to receive a training image; a mask generator module configured to; generate candidate segment masks based on the training image received by the training supervisor module; and rank the candidate segment masks to generate ranked candidate segment masks; a mask selector module configured to select a set of the ranked candidate segment masks to generate a set of ranked candidate segment masks; and a neural network updater module configured to; select one of the set of ranked candidate segment masks to train the neural network; and train the neural network. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8)
-
-
9. A method, comprising:
-
receiving an input image; generating candidate segment masks based on the input image; and performing an initial training operation comprising; ranking the candidate segment masks to generate ranked candidate segment masks; selecting a set of the ranked candidate segment masks to generate a set of the ranked candidate segment masks; selecting one mask of the set of the ranked candidate segment masks; and training a neural network by applying the selected mask of the set of the ranked candidate segment masks to the network. - View Dependent Claims (10, 11, 12, 13, 14, 15, 16, 17)
-
-
18. A computer-readable medium having computer-executable instructions thereupon that, when executed by a computer, cause the computer to:
-
receive an input image comprising a ground-truth bounding box; generate candidate segment masks based on the input image; and perform an training operation comprising ranking the candidate segment masks to generate ranked candidate segment masks, selecting a set of the ranked candidate segment masks to generate a set of the ranked candidate segment masks, randomly selecting one mask of the set of the ranked candidate segment masks, and training a neural network by applying the selected mask of the set of the ranked candidate segment masks to the network. - View Dependent Claims (19, 20)
-
Specification