Systems and methods for automatically clustering similar objects in images
A method for automatically clustering similar objects in images includes accessing a plurality of video frames of a video. The method further includes identifying a plurality of objects from the plurality of video frames. The method further includes generating a plurality of composite vectors for the plurality of objects by: generating a plurality of vectors for each particular object of the plurality of objects extracted from the plurality of video frames; and generating a particular composite vector for each particular object by combining the plurality of vectors for the particular object. The method further includes determining, using the composite vectors for the plurality of objects, a plurality of similar objects. The method further includes displaying images of one or more of the plurality of similar objects in a graphical user interface.
1 . A system comprising:
a camera;
one or more memory units; and
one or more computer processors communicatively coupled to the one or more memory units and configured to perform operations comprising:
access a plurality of video frames of a video;
identify a plurality of objects from the plurality of video frames by:
generating, using a segmentation algorithm, a plurality of masks, each mask comprising a set of neighboring pixels that are determined to be related; and
extracting the plurality of objects from the plurality of video frames based on the generated plurality of masks;
generate a plurality of composite vectors for the plurality of objects by:
generating a plurality of vectors for each particular object of the plurality of objects extracted from the plurality of video frames; and
generating a particular composite vector for each particular object by combining the plurality of vectors for the particular object;
determine, using the composite vectors for the plurality of objects, a plurality of similar objects; and
display images of one or more of the plurality of similar objects in a graphical user interface.
2 . The system of claim 1 , wherein the plurality of vectors that are generated for each particular object of the plurality of objects extracted from the plurality of video frames comprises:
an appearance vector that is a mathematical representation of the appearance of the particular object;
a behavior vector that is a mathematical representation of the behavior or movement of the particular object; and
a shape vector that is a mathematical representation of the size or shape of the particular object.
3 . The system of claim 1 , the operations further comprising displaying a user-editable label for the plurality of similar objects in the graphical user interface.
4 . The system of claim 1 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of objects using the composite vectors for the plurality of objects, each similarity score denoting the similarity between two of the plurality of objects.
5 . The system of claim 4 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
accessing a user-selection of a selected object of the plurality of objects;
calculating similarity scores between the selected object and other objects of the plurality of objects; and
selecting the other objects of the plurality of objects whose calculated similarity scores with the selected object meets a predetermined similarity value.
6 . The system of claim 4 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
automatically calculating similarity scores between each particular object and every other object of the plurality of objects; and
clustering the plurality of objects based on the calculated similarity scores.
7 . A method by a computing system for automatically clustering similar objects in images, the method comprising:
accessing a plurality of video frames of a video;
identifying a plurality of objects from the plurality of video frames by:
generating, using a segmentation algorithm, a plurality of masks, each mask comprising a set of neighboring pixels that are determined to be related; and
extracting the plurality of objects from the plurality of video frames based on the generated plurality of masks;
generating a plurality of composite vectors for the plurality of objects by:
generating a plurality of vectors for each particular object of the plurality of objects extracted from the plurality of video frames; and
generating a particular composite vector for each particular object by combining the plurality of vectors for the particular object;
determining, using the composite vectors for the plurality of objects, a plurality of similar objects; and
displaying images of one or more of the plurality of similar objects in a graphical user interface.
8 . The method of claim 7 , wherein the plurality of vectors that are generated for each particular object of the plurality of objects extracted from the plurality of video frames comprises:
an appearance vector that is a mathematical representation of the appearance of the particular object;
a behavior vector that is a mathematical representation of the behavior or movement of the particular object; and
a shape vector that is a mathematical representation of the size or shape of the particular object.
9 . The method of claim 7 , further comprising displaying a user-editable label for the plurality of similar objects in the graphical user interface.
10 . The method of claim 7 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of objects using the composite vectors for the plurality of objects, each similarity score denoting the similarity between two of the plurality of objects.
11 . The method of claim 10 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
accessing a user-selection of a selected object of the plurality of objects;
calculating similarity scores between the selected object and other objects of the plurality of objects; and
selecting the other objects of the plurality of objects whose calculated similarity scores with the selected object meets a predetermined similarity value.
12 . The method of claim 10 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
automatically calculating similarity scores between each particular object and every other object of the plurality of objects; and
clustering the plurality of objects based on the calculated similarity scores.
13 . One or more computer-readable non-transitory storage media embodying instructions that, when executed by a processor, cause the processor to perform operations comprising:
access a plurality of video frames of a video;
identify a plurality of objects from the plurality of video frames by:
generating, using a segmentation algorithm, a plurality of masks, each mask comprising a set of neighboring pixels that are determined to be related; and
extracting the plurality of objects from the plurality of video frames based on the generated plurality of masks;
generate a plurality of composite vectors for the plurality of objects by:
generating a plurality of vectors for each particular object of the plurality of objects extracted from the plurality of video frames; and
generating a particular composite vector for each particular object by combining the plurality of vectors for the particular object;
determine, using the composite vectors for the plurality of objects, a plurality of similar objects; and
display images of one or more of the plurality of similar objects in a graphical user interface.
14 . The one or more computer-readable non-transitory storage media of claim 13 , wherein the plurality of vectors that are generated for each particular object of the plurality of objects extracted from the plurality of video frames comprises:
an appearance vector that is a mathematical representation of the appearance of the particular object;
a behavior vector that is a mathematical representation of the behavior or movement of the particular object; and
a shape vector that is a mathematical representation of the size or shape of the particular object.
15 . The one or more computer-readable non-transitory storage media of claim 13 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises calculating a plurality of similarity scores for the plurality of objects using the composite vectors for the plurality of objects, each similarity score denoting the similarity between two of the plurality of objects.
16 . The one or more computer-readable non-transitory storage media of claim 15 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
accessing a user-selection of a selected object of the plurality of objects;
calculating similarity scores between the selected object and other objects of the plurality of objects; and
selecting the other objects of the plurality of objects whose calculated similarity scores with the selected object meets a predetermined similarity value.
17 . The one or more computer-readable non-transitory storage media of claim 15 , wherein determining, using the composite vectors for the plurality of objects, the plurality of similar objects comprises:
automatically calculating similarity scores between each particular object and every other object of the plurality of objects; and
clustering the plurality of objects based on the calculated similarity scores.