Training of machine learning models for video analytics

1 . A method implemented in a management node, the method comprising:

receiving training data for a machine learning (ML) model, the training data comprising a plurality of video segments, each video segment of the plurality of video segments comprising a first subset of a plurality of video frames and a second subset of the plurality of video frames;

selecting the first subset of the plurality of video frames of each video segment for training a first machine learning (ML) model, the selection being based on a first selection criteria that comprises rules based on a subsegment characteristic, the subsegment characteristic comprising:

a weight of a pre-weighed subsegment; or

association with an object present in one or more of previously selected subsegments and in an additional subsegment, the object in the additional subsegment having at least one modified characteristic;

training the first ML model using the first subset of the plurality of video frames for each video segment that were selected based on the first selection criteria, the second subset of the plurality of frames not being used for training the first ML model;

generating a first accuracy metric for the first ML model;

selecting the second subset of the plurality of video frames of each video segment for training a second ML model, the selection being based on a second selection criteria;

training the second ML model using the second subset of the plurality of video frames for each video segment that were selected based on the second selection criteria, the first subset of the plurality of video frames not being used for training the second ML model;

generating a second accuracy metric for the second ML model; and

selecting one of the first ML model or the second ML model for use in a system based on a comparison of the first accuracy metric and the second accuracy metric.

2 . A method implemented in a management node, the method comprising:

receiving a video segment comprising a first plurality of subsegments and a second plurality of subsegments;

selecting the first plurality of subsegments of the video segment for training a machine learning (ML) model, the selection being based on a selection criteria that comprises rules based on a subsegment characteristic, the subsegment characteristic comprising:

a weight of a pre-weighed subsegment; or

association with an object present in one or more of previously selected subsegments and in an additional subsegment, the object in the additional subsegment having at least one modified characteristic;

training the ML model using the first plurality of subsegments that were selected based on the selection criteria, the second plurality of subsegments of the video segment not being used for training the ML model; and

after training the ML model, testing the trained ML model based on an accuracy threshold.

3 . The method of claim 2 , wherein the rules are further based on a video segment characteristic comprising a length of the video segment.

4 . The method of claim 2 , wherein the rules are further based on a system characteristic comprising a computing parameter of the management node.

5 . The method of claim 2 , further comprising:

selecting a plurality of additional subsegments of the video segment based on an additional selection criteria, the plurality of additional subsegments being different from the first plurality of subsegments and the second plurality of subsegments; and

training an additional ML model using the plurality of additional subsegments selected based on the additional selection criteria.

6 . The method of claim 5 , further comprising testing the additional ML model based on the accuracy threshold.

7 . The method of claim 6 , further comprising selecting one of the ML model or the additional ML model based on a result of testing the ML model and the additional ML model based on the accuracy threshold.

8 . The method of claim 5 , further comprising:

generating an accuracy metric for the ML model;

generating an additional accuracy metric for the additional ML model; and

selecting one of the ML model or the additional ML model for deployment based on the accuracy metric and the additional accuracy metric.

9 . The method of claim 2 , further comprising:

receiving an additional video segment;

selecting a plurality of additional video subsegments of the additional video segment based on the selection criteria;

generating a grouping of the first plurality of subsegments and the plurality of additional video subsegments; and

training the ML model using the grouping of the first plurality of subsegments and the plurality of additional video subsegments.

10 . A management node comprising:

at least one processor; and

at least one memory storing instructions that, when executed, cause the at least one processor to:

receive a video segment comprising a first plurality of subsegments and a second plurality of subsegments;

select the first plurality of subsegments of the video segment for training a machine learning (ML) model, the selection being based on a selection criteria that comprises rules based on a subsegment characteristic, the subsegment characteristic comprising:

a weight of a pre-weighed subsegment; or

association with an object present in one or more of previously selected subsegments and in an additional subsegment, the object in the additional subsegment having at least one modified characteristic;

train the ML model using the first plurality of subsegments that were selected based on the selection criteria, the second plurality of subsegments of the video segment not being used for training the ML model; and

after training the ML model, test the trained ML model based on an accuracy threshold.

11 . The management node of claim 10 , wherein the rules are further based on a video segment characteristic comprising a length of the video segment.

12 . The management node of claim 10 , wherein the rules are further based on a system characteristic comprising a computing parameter of the management node.

13 . The management node of claim 10 , wherein the at least one memory stores computer the instructions that, when executed by the at least one processor, further cause the at least one processor to:

select a plurality of additional subsegments of the video segment based on an additional selection criteria, the plurality of additional subsegments being different from the first plurality of subsegments and the second plurality of subsegments; and

train an additional ML model using the plurality of additional subsegments selected based on the additional selection criteria.

14 . The management node of claim 13 , wherein the at least one memory stores computer the instructions that, when executed by the at least one processor, further cause the at least one processor to test the additional ML model based on the accuracy threshold.

15 . The management node of claim 14 , wherein the at least one memory stores computer the instructions that, when executed by the at least one processor, further cause the at least one processor to select one of the ML model or the additional ML model based on a result of testing the ML model and the additional ML model based on the accuracy threshold.

16 . The management node of claim 13 , wherein the at least one memory stores computer the instructions that, when executed by the at least one processor, further cause the at least one processor to:

generate an accuracy metric for the ML model;

generate an additional accuracy metric for the additional ML model; and

select one of the ML model or the additional ML model for deployment based on the accuracy metric and the additional accuracy metric.