Filtering detected objects from an object recognition index according to extracted features

Patent metadata

Field	Value
Publication number	US-2022139063-A1
Application number	US-202117525413-A
Country	US
Kind code	A1
Filing date	Nov 12, 2021
Priority date	Sep 6, 2018
Publication date	May 5, 2022
Grant date	—

Abstract

Objects detected in data may be filtered from an object recognition index. Data for object detection may be received. An object detection technique may be applied to the data to detect an object. If the object does not satisfy indexing criteria for the object recognition index, then the detected object may be excluded from the object recognition index.

First claim

What is claimed is:

1 . A system, comprising:
at least one processor; and a memory, storing program instructions that when executed by the at least one processor cause the at least one processor to:
receive data to add objects detected in the data to an object recognition index;
detect an object in the data according to an object detection technique;
extract one or more features for the detected object;
evaluate the one or more features of the detected object according to one or more indexing criteria to determine whether to exclude objects from the object recognition index; and
responsive to a determination that the detected object does not satisfy the one or more indexing techniques, exclude the detected object detected from the object recognition index according to a determination that the one or more features of the object do not satisfy the one or more indexing criteria.

2 . The system of claim 1 , wherein the program instructions further cause the at least one processor to return the representation of the one or more features of the detected object responsive to a request for objects not included in the object recognition index.

3 . The system of claim 1 , wherein to extract the one or more features for the detected object, the program instructions cause the at least one processor to:
apply a convolutional neural network trained to indicate similarity between detected objects to the detected object in order to determine a feature vector that represents the one or more features of the detected object; or apply the convolutional neural network to determine one or more intermediate features in order to determine the one or more features as domain specific attributes of the detected object.

4 . The system of claim 1 , wherein the at least one processor and the memory are implemented as part of an object recognition service of a provider network that detects human faces in images, wherein the data is image data, wherein the object recognition index is hosted in the provider network, and wherein the detection, evaluation, extraction, and exclusion are performed responsive to a request to index the image data as part of the object recognition index.

5 . A method, comprising:
applying, by one or more computing devices, an object detection technique on data to detect an object within the data to be considered for inclusion in an object recognition index; evaluating, by the one or more computing devices, one or more features of the detected object determined as part of the application of the object detection technique with respect to one or more indexing criteria to exclude objects from the object recognition index that do not satisfy the one or more indexing criteria; and excluding, by the one or more computing devices, the detected object detected from the object recognition index according to a determination that the one or more features of the object do not satisfy the one or more indexing criteria.

6 . The method of claim 5 , further comprising receiving, by the one or more computing devices, a request to index objects within the data, the applying, the evaluating, and the excluding are performed responsive to the request.

7 . The method of claim 6 , further comprising:
responsive to the request, obtaining the data from a data store indicated by the request.

8 . The method of claim 6 , wherein the request includes a parameter that requests application of the one or more indexing criteria when considering objects detected in the data for inclusion in the object recognition index.

9 . The method of claim 5 , further comprising:
wherein the applying of the object detection technique on the data detects a second object within the data; evaluating, by the one or more computing devices, one or more features of the second detected object determined as part of the application of the object detection technique with respect to the one or more indexing criteria; and including, by the one or more computing devices, the second detected object in the object recognition index according to a determination that the one or more features of the second detected object satisfy the one or more indexing criteria.

10 . The method of claim 5 , further comprising:
after an update to the object detection technique:
obtaining, by the one or more computing devices, respective features for true and falsely detected objects according to the updated object detection technique applied to a plurality of data;
training, by the one or more computing devices, a predictive model according to the obtained features to identify respective feature values that maximize a prediction of one of the true detected objects upon application of the predictive model to the respective features of the one true detected object; and
updating, by the one or more computing devices, the one or more indexing criteria based on the respective features identified by the predictive model that maximize the prediction of true detected objects.

11 . The method of claim 5 , further comprising returning, by the one or more computing devices, a response via an interface indicating that the object was not included in the object recognition index and an indication of at least one of the respective features of the object that failed to satisfy the one or more indexing criteria.

12 . The method of claim 5 , wherein the object detection technique is a natural language processing technique.

13 . The method of claim 5 , further comprising:
applying, by the one or more computing devices, a second object detection technique on second data to detect an object within the second data to be considered for inclusion in an second object recognition index; evaluating, by the one or more computing devices, one or more features of the detected object determined as part of the application of the second object detection technique with respect to one or more other indexing criteria to exclude objects from the second object recognition index that do not satisfy the one or more indexing criteria, wherein the one or more other indexing criteria are different than the one or more indexing criteria; and excluding, by the one or more computing devices, the detected object detected from the second object recognition index according to a determination that the one or more features of the object do not satisfy the one or more other indexing criteria.

14 . One or more non-transitory, computer-readable storage media, storing program instructions that when executed on or across one or more computing devices cause the one or more computing devices to implement:
receiving image data to add objects detected in the image data to an object recognition index; applying an object detection technique on the image data to detect an object within the image data; evaluating one or more features of the detected object determined as part of the application of the object detection technique with respect to one or more indexing criteria to exclude objects from the object recognition index that do not satisfy the one or more indexing criteria; and excluding the detected object detected from the object recognition index according to a determination that the one or more features of the object do not satisfy the one or more indexing criteria.

15 . The one or more non-transitory, computer-readable storage media of claim 14 , wherein the program instructions cause the one or more computing devices to further implement:
wherein the applying of the object detection technique on the image data detects a second object within the image data; evaluating, by the one or more computi

Assignees

• Amazon Tech Inc

Inventors

• Singh Kunwar Yashraj
• Johnson Keith Young
• Bhadauria Vivek
• Flynn Sean R
• Du Binglei
• Thomas Dylan C
• Manohar Vasant
• Hedley Jonathan
• Xia Wei

Classifications

• G06V20/647 Primary

  by matching two-dimensional images to three-dimensional objects · CPC title

• G06V10/7788

  the supervisor being a human, e.g. interactive learning with a human teacher · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06V10/771

  Feature selection, e.g. selecting representative features from a multi-dimensional feature space · CPC title