Method and Apparatus for Improved Training of Object Detecting System
First Claim
Patent Images
1. A system for generating a computer vision solution for a specified vision task, the system comprising:
- an input configured to receive user-defined limitations, wherein at least one of said user-defined limitations is associated with a sub-set of predefined descriptors within a library of predefined descriptors, each of said descriptors being associated with a respective set of image constraints; and
a processor configured to;
for each user-defined limitation that is associated with said library of predefined descriptors, selecting an available descriptor within its associated sub-set of predefined descriptors that is most consistent with all user-defined limitations;
defining a working image library of sample images based on the user-defined limitations and the sets of image constraints associated with the selected descriptors;
defining an evaluation set of sample images from said working image library;
defining a vision solution candidate based on the user-defined limitations and the selected descriptors;
tuning said vision solution candidate, including modifying the selected descriptors according to the specified vision task;
evaluating the tuned vision solution candidate using said evaluation set of sample images and user-provided limitations, including accuracy and speed limitations;
using evaluation feedback for additional tuning of the selected solution candidate until the accuracy and speed limitations are met or a maximum of additional tuning is reached.
2 Assignments
0 Petitions
Accused Products
Abstract
An adequate solution for computer vision applications is arrived at more efficiently and, with more automation, enables users with limited or no special image processing and pattern recognition knowledge to create reliable vision systems for their applications. Computer rendering of CAD models is used to automate the dataset acquisition process and labeling process. In order to speed up the training data preparation while maintaining the data quality, a number of processed samples are generated from one or a few seed images.
60 Citations
20 Claims
-
1. A system for generating a computer vision solution for a specified vision task, the system comprising:
-
an input configured to receive user-defined limitations, wherein at least one of said user-defined limitations is associated with a sub-set of predefined descriptors within a library of predefined descriptors, each of said descriptors being associated with a respective set of image constraints; and a processor configured to; for each user-defined limitation that is associated with said library of predefined descriptors, selecting an available descriptor within its associated sub-set of predefined descriptors that is most consistent with all user-defined limitations; defining a working image library of sample images based on the user-defined limitations and the sets of image constraints associated with the selected descriptors; defining an evaluation set of sample images from said working image library; defining a vision solution candidate based on the user-defined limitations and the selected descriptors; tuning said vision solution candidate, including modifying the selected descriptors according to the specified vision task; evaluating the tuned vision solution candidate using said evaluation set of sample images and user-provided limitations, including accuracy and speed limitations; using evaluation feedback for additional tuning of the selected solution candidate until the accuracy and speed limitations are met or a maximum of additional tuning is reached. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
-
-
20. One or more tangible, non-transitory computer-readable media embodying instructions executable by a compute to perform a method for automatically generating a computer vision solution, the method comprising:
-
receiving user input for one or more descriptors; creating a training set of object images and an evaluation set of object images; selecting a vision solution candidate that provides a best match to the user input from a predefined solutions library; training the selected solution candidate using the training set; applying the selected solution candidate to the evaluation set; evaluating the selected solution candidate solution using the user input; and using evaluation feedback for additional training of the selected solution candidate until accuracy and speed requirements are met or a maximum of additional training is reached.
-
Specification