Method and server for recognizing and determining object in image
Assigneedoinglab Corp.
InventorsJae Yeol Kim, Dae Han Kim, Hyun Suk Lee
Life of this patent
Estimating the term…
Abstract
A method in which a server recognizes and determines an object in an image may include calculating a feature map including an object feature point within the image by learning the image based on a first model, calculating compression feature information by compressing the feature map based on a first filter, inputting the compression feature information for each at least one layer and calculating respective prediction information by learning the compression feature information for each layer based on a second model, and calculating prediction result information by determining whether the respective prediction information sequentially exceeds a predetermined threshold value in order from a top layer.
Claims13
A method in which a server recognizes and determines an object in an image, the method comprising:
- calculating a feature map including an object feature point within the image by learning the image based on a first model;
- calculating compression feature information by compressing the feature map based on a first filter;
- inputting the compression feature information for each at least one layer and calculating respective prediction information by learning the compression feature information for each layer based on a second model;
- and
- calculating prediction result information by determining whether the respective prediction information sequentially exceeds a predetermined threshold value in order from a top layer,
- wherein the layer includes at least one of first to kth layers, is implemented sequentially such that a category of an upper layer includes a category of a lower layer, and includes meta information about an immediately-upper layer and an immediately-lower layer for the each layer.
A server recognizing and determining an object in an image, the server comprising:
- an image obtaining unit configured to obtain the image;
- a memory;
- and
- a processor is configured to:
- calculate a feature map including an object feature point within the image by learning the image based on a first model;
- calculate compression feature information by compressing the feature map based on a first filter;
- input the compression feature information for each at least one layer and calculate respective prediction information by learning the compression feature information for each layer based on a second model;
- and
- calculate prediction result information by determining whether the respective prediction information sequentially exceeds a predetermined threshold value in order from a top layer,
- wherein the layer includes at least one of first to kth layers, is implemented sequentially such that a category of an upper layer includes a category of a lower layer, and includes meta information about an immediately-upper layer and an immediately-lower layer for the each layer.
Dependencies are inferred from the claim text.