Content type detection in videos using multiple classifiers

1. A method comprising:

receiving, by a computing device, a set of frames from a video at a first classifier;

classifying, by the computing device, the set of frames with a set of classification scores indicating a confidence that a frame contains end credit content using the first classifier, the first classifier using a first model that classifies content from the set of frames;

after performing the classifying by the first classifier, performing:

receiving, by the computing device, at least a portion of the set of classification scores for at least a portion of the set of frames from the first classifier;

adjusting, by the computing device, a classification score in the set of classification scores for a frame to another classification score using one or more classification scores from one or more of the at least the portion of the set of frames that are considered to be neighboring frames to the frame using a second classifier, the second classifier using a second model that classifies classification scores from the first classifier and not content of the at least the portion of the set of frames that was used by the first classifier; and

selecting, by the computing device, a boundary point in the set of frames between a first frame in the set of frames that is considered to not include end credit content and a second frame in the set of frames that is considered to include end credit content using the at least the portion of the set of classification scores with the classification score being replaced with the adjusted classification score.

2. The method of claim 1 , wherein classifying the set of frames using the first classifier comprises:

receiving a frame in the set of frames;

analyzing content in the frame; and

selecting between a first classification node that the frame includes the end credits content and a second classification node that the frame does not include the end credits content.

3. The method of claim 2 , wherein the classification score indicates the confidence in selecting one of the first classification and the second classification.

4. The method of claim 2 , wherein the first classifier is configured with output nodes that output a first classification score for the first classification node and a second classification score for the second classification node.

5. The method of claim 1 , further comprising:

inputting the at least the portion of the set of classification scores into the second model; and

adjusting at least a portion of the set of classification scores to different classification scores based on the one or more classification scores of the one or more neighboring frames in the set of frames.

6. The method of claim 1 , wherein adjusting the classification score using the second classifier comprises:

inputting the classification score into the second model; and

adjusting the classification score based on secondary information for the frame or the neighboring frames.

7. The method of claim 1 , wherein adjusting the classification score using the second classifier comprises:

changing a classification of the frame from including end credit content to not including end credit content or from not including end credit content to including end credit content based on the one or more classification scores of one or more of the at least the portion of the set of frames.

8. The method of claim 1 , wherein selecting the boundary point comprises:

selecting a prospective boundary point;

calculating a left window score based on classification scores for at least a portion of frames before the prospective boundary point;

calculating a right window score based on classification scores for at least a portion of frames after the prospective boundary point; and

calculating a boundary score based on the left window score and the right window score.

9. The method of claim 8 , further comprising:

continuing to select different prospective boundary points;

calculating the left window score and the right window score based on the different boundary points; and

calculating different boundary scores for the different prospective boundary points.

10. The method of claim 9 , further comprising:

selecting one of the different boundary scores for the boundary point.

11. The method of claim 10 , wherein the selected one of the different boundary scores is a maximum score out of the different boundary scores.

12. The method of claim 1 , further comprising:

extracting the set of frames from the video based on a point in the video.

13. The method of claim 12 , wherein extracting the set of frames comprises:

extracting frames after a time in the video to form the set of frames.

14. The method of claim 1 , wherein:

the video is received during a live broadcast, and

the boundary point is selected during the live broadcast.

15. The method of claim 1 , further comprising:

selecting neighboring frames to the frame; and

connecting the classification score in the set of classification scores for the frame with the one or more classification scores from the one or more of the at least the portion of the set of frames that are considered to be neighboring frames to the frame.

16. A non-transitory computer-readable storage medium containing instructions, that when executed, control a computer system to be configured for:

receiving a set of frames from a video at a first classifier;

classifying the set of frames with a set of classification scores indicating a confidence that a frame contains end credit content using the first classifier, the first classifier using a first model that classifies content from the set of frames;

after performing the classifying by the first classifier, performing:

receiving at least a portion of the set of classification scores for at least a portion of the set of frames from the first classifier;

adjusting a classification score in the set of classification scores for a frame to another classification score using one or more classification scores from one or more of the at least the portion of the set of frames that are considered to be neighboring frames to the frame using a second classifier, the second classifier using a second model that classifies classification scores from the first classifier and not content of the at least the portion of the set of frames that was used by the first classifier; and

selecting a boundary point in the set of frames between a first frame in the set of frames that is considered to not include end credit content and a second frame in the set of frames that is considered to include end credit content using the at least the portion of the set of classification scores with the classification score being replaced with the adjusted classification score.

17. The non-transitory computer-readable storage medium of claim 16 , wherein classifying the set of frames using the first classifier comprises:

receiving a frame in the set of frames;

analyzing content in the frame; and

selecting between a first classification node that the frame includes the end credits content and a second classification node that the frame does not include the end credits content.

18. The non-transitory computer-readable storage medium of claim 16 , further configured for:

inputting the at least a portion of the set of classification scores into the second model; and

adjusting at least a portion of the set of classification scores to different classification scores based on the one or more classification scores of the one or more neighboring frames in the set of frames.

19. The non-transitory computer-readable storage medium of claim 16 , wherein adjusting the classification scores using the second classifier comprises:

changing a classification of the frame from including end credit content to not including end credit content or from not including end credit content to including end credit content based on the classification scores of the neighboring frames in the set of frames.

20. The non-transitory computer-readable storage medium of claim 16 , wherein selecting the boundary point comprises:

selecting a prospective boundary point;

calculating a left window score based on classification scores for at least a portion of frames before the prospective boundary point;

calculating a right window score based on classification scores for at least a portion of frames after the prospective boundary point; and

calculating a boundary score based on the left window score and the right window score.

21. An apparatus comprising:

one or more computer processors; and

a non-transitory computer-readable storage medium comprising instructions, that when executed, control the one or more computer processors to be configured for:

receiving a set of frames from a video at a first classifier;

classifying the set of frames with a set of classification scores indicating a confidence that a frame contains end credit content using the first classifier, the first classifier using a first model that classifies content from the set of frames;

after performing the classifying by the first classifier, performing:

receiving at least a portion of the set of classification scores for at least a portion of the set of frames from the first classifier;

adjusting a classification score in the set of classification scores for a frame to another classification score using one or more classification scores from one or more of the at least the portion of the set of frames that are considered to be neighboring frames to the frame using a second classifier, the second classifier using a second model that classifies classification scores from the first classifier and not content of the at least the portion of the set of frames that was used by the first classifier; and

selecting a boundary point in the set of frames between a first frame in the set of frames that is considered to not include end credit content and a second frame in the set of frames that is considered to include end credit content using the at least the portion of the set of classification scores with the classification score being replaced with the adjusted classification score.