Method for object detection using knowledge distillation
AssigneeCortica, Ltd.
InventorsIgal Raichelgauz, Eli Passov
Life of this patent
Estimating the term…
Abstract
A method that may include training a student ODNN to mimic a teacher ODNN. The training may include calculating a teacher student detection loss that is based on a pre-bounding-box output of the teacher ODNN. The pre-bounding-box output of the teacher ODNN is a function of pre-bounding-box outputs of different ODNNs that belong to the teacher ODNN. The method may also include detecting one or more objects in an image, by feeding the image to the trained student ODNN; outputting by the trained student ODNN a student pre-bounding-box output; and calculating one or more bounding boxes based on the student pre-bounding-box output.
Claims16
A method for object detection, the method comprises:
- calculating, by a teacher object detection neural network (ODNN), weights to be applied during a calculation of a weighted sum;
- training a student object detection neural network (ODNN) to mimic a teacher ODNN;
- wherein the training comprises calculating a teacher student detection loss that is based on a pre-bounding-box output of the teacher ODNN that is a weighted sum of pre-bounding-box outputs of different ODNNs that belong to the teacher ODNN;
- and
- detecting one or more objects in an image;
- wherein the detecting comprises:
- feeding the image to the trained student ODNN;
- outputting by the trained student ODNN a student pre-bounding-box output;
- and
- calculating one or more bounding boxes based on the student pre-bounding-box output.
A non-transitory computer readable medium that stores instructions for:
- calculating, by a teacher object detection neural network (ODNN), weights to be applied during a calculation of a weighted sum;
- training a student ODNN to mimic the teacher ODNN;
- wherein the training comprises calculating a teacher student detection loss that is based on a pre-bounding-box output of the teacher ODNN that is a weighted sum of pre-bounding-box outputs of different ODNNs that belong to the teacher ODNN;
- and
- detecting one or more objects in an image;
- wherein the detecting comprises:
- feeding the image to the trained student ODNN;
- outputting by the trained student ODNN a student pre-bounding-box output;
- and
- calculating one or more bounding boxes based on the student pre-bounding-box output.
Dependencies are inferred from the claim text.