Removing independent noise using DeepInterpolation
A facility for transforming a subject data item sequence is described. The facility accesses a trained relationship model. For each of a plurality of subject items of the subject data item sequence, the facility: selects a first contiguous series of items of the subject data item sequence immediately before the subject data item; selects a second contiguous series of items of the subject data item sequence immediately after the subject data item; and applies the trained relationship model to the selected first and second contiguous series of data items to obtain a denoised version of the subject data item. The facility then assembles the obtained denoised subject data items into a denoised data item sequence.
1. A method in a computing system for generating a denoised video sequence from a source video sequence, comprising:
accessing a set of training video sequences;
among the training video sequences, selecting a plurality of contiguous series of frames;
for each of the plurality of selected contiguous series of frames, defining a training observation in which a central frame in the contiguous series of frames is the dependent variable, and the frames of the contiguous series of frames other than the central frame are independent variables;
training a machine learning model using the defined training observations;
for each of a plurality of subject frames of the source video sequence:
selecting a first contiguous series of frames of the source video sequence immediately before the subject frame;
selecting a second contiguous series of frames of the source video sequence immediately after the subject frame; and
applying the trained machine learning model to the selected first and second contiguous series of frames, to the exclusion of the subject frame, to obtain a denoised version of the subject frame; and
assembling the obtained denoised subject frame versions into the denoised video sequence.
2. The method of claim 1 , further comprising storing the assembled denoised video sequence.
3. The method of claim 1 , further comprising causing the assembled denoised video sequence to be displayed.
4. The method of claim 1 wherein the set of training video sequences contains only the source sequence.
5. The method of claim 1 wherein the set of training video sequences is a plurality of video sequences that includes the source sequence.
6. The method of claim 1 wherein the set of training video sequences is a plurality of video sequences that excludes the source sequence.
7. A computing system for generating a denoised video sequence from a source video sequence, the computing system being configured to:
access a set of training video sequences;
among the training video sequences, select a plurality of contiguous series of frames;
for each of the plurality of selected contiguous series of frames, define a training observation in which a central frame in the contiguous series of frames is the dependent variable, and the frames of the contiguous series of frames other than the central frame are independent variables;
train a machine learning model using the defined training observations;
for each of a plurality of subject frames of the source video sequence:
select a first contiguous series of frames of the source video sequence immediately before the subject frame;
select a second contiguous series of frames of the source video sequence immediately after the subject frame; and
apply the trained machine learning model to the selected first and second contiguous series of frames, to the exclusion of the subject frame, to obtain a denoised version of the subject frame; and
assemble the obtained denoised subject frame versions into the denoised video sequence.
8. The computing system of claim 7 , wherein the computing system is further configured to store the assembled denoised video sequence.
9. The computing system of claim 7 , wherein the computing system is further configured to cause the assembled denoised video sequence to be displayed.
10. The computing system of claim 7 , wherein the set of training video sequences contains only the source sequence.
11. The computing system of claim 7 , wherein the set of training video sequences is a plurality of video sequences that includes the source sequence.
12. The computing system of claim 7 , wherein set of training video sequences is a plurality of video sequences that excludes the source sequence.
13. One or more instances of computer-readable media, the one or more instances of computer-readable media not constituting a transitory propagating data signal and collectively having contents configured to cause a computing system for generating a denoised video sequence from a source video sequence to perform a method, the method comprising:
accessing a set of training video sequences;
among the training video sequences, selecting a plurality of contiguous series of frames;
for each of the plurality of selected contiguous series of frames, defining a training observation in which a central frame in the contiguous series of frames is the dependent variable, and the frames of the contiguous series of frames other than the central frame are independent variables;
training a machine learning model using the defined training observations;
for each of a plurality of subject frames of the source video sequence:
selecting a first contiguous series of frames of the source video sequence immediately before the subject frame;
selecting a second contiguous series of frames of the source video sequence immediately after the subject frame; and
applying the trained machine learning model to the selected first and second contiguous series of frames, to the exclusion of the subject frame, to obtain a denoised version of the subject frame; and
assembling the obtained denoised subject frame versions into the denoised video sequence.
14. The one or more instances of computer-readable media of claim 13 , wherein the method further comprises storing the assembled denoised video sequence.
15. The one or more instances of computer-readable media of claim 13 , wherein the method further comprises causing the assembled denoised video sequence to be displayed.
16. The one or more instances of computer-readable media of claim 13 , wherein the set of training video sequences contains only the source sequence.
17. The one or more instances of computer-readable media of claim 13 , wherein the set of training video sequences is a plurality of video sequences that includes the source sequence.
18. The one or more instances of computer-readable media of claim 13 , wherein the set of training video sequences is a plurality of video sequences that excludes the source sequence.