Use of a saliency map to train a colorization ANN
Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device could receive an input image comprising a plurality of channels and determine a saliency map for the input image. The computing device could also establish at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels. Further, the computing device could train an ANN to predict one or more output channels from the one or more training channels, where the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and where the difference is computationally biased based on values from the saliency map.
1. A computer-implemented method comprising:
receiving, at a computing device, an input image comprising a plurality of channels;
determining, by the computing device, a saliency map for the input image;
establishing, by the computing device, at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels;
training, by the computing device, an artificial neural network (ANN) to predict one or more output channels from the training channel, wherein the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and wherein the difference is computationally biased based on values from the saliency map; and
providing, using the computing device, the trained ANN.
2. The computer-implemented method of claim 1 , wherein the difference between the one or more output channels and the one or more ground truth channels comprises a pixel-wise difference between the one or more output channels and the one or more ground truth channels.
3. The computer-implemented method of claim 2 , wherein the saliency map comprises respective values for each pixel in the input image, and wherein the computational bias involves scaling the pixel-wise difference between the one or more output channels and the one or more ground truth channels by the respective values.
4. The computer-implemented method of claim 1 , wherein the input image is associated with at least one object class, and wherein the saliency map is determined with respect to the at least one object class.
5. The computer-implemented method of claim 4 , wherein the at least one object class corresponds to at least one foreground object in the input image.
6. The computer-implemented method of claim 1 , wherein the input image is part of a plurality of input images, and wherein the receiving, determining, establishing, and training are also applied to each of the plurality of input images.
7. The computer-implemented method of claim 1 , wherein the providing comprises providing the trained ANN to a printing device.
8. The computer-implemented method of claim 1 , wherein the training channel corresponds to a lightness channel, wherein the one or more ground truth channels correspond to one or more color channels, and wherein the training involves predicting one or more output color channels from the lightness channel.
9. A computing device, comprising:
one or more processors; and
non-transitory data storage storing at least computer-readable instructions that, when executed by the one or more processors, cause the computing device to perform operations comprising:
receiving an input image comprising a plurality of channels;
determining a saliency map for the input image;
establishing at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels;
training an artificial neural network (ANN) to predict one or more output channels from the training channel, wherein the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and wherein the difference is computationally biased based on values from the saliency map; and
providing the trained ANN.
10. The computing device of claim 9 , wherein the difference between the one or more output channels and the one or more ground truth channels comprises a pixel-wise difference between the one or more output channels and the one or more ground truth channels.
11. The computing device of claim 10 , wherein the saliency map comprises respective values for each pixel in the input image, and wherein the computational bias involves scaling the pixel-wise difference between the one or more output channels and the one or more ground truth channels by the respective values.
12. The computing device of claim 9 , wherein the input image is associated with at least one object class, and wherein the saliency map is determined with respect to the at least one object class.
13. The computing device of claim 12 , wherein the at least one object class corresponds to at least one foreground object in the input image.
14. The computing device of claim 9 , wherein the input image is part of a plurality of input images, and wherein the receiving, determining, establishing, and training are also applied to each of the plurality of input images.
15. The computing device of claim 9 , wherein the providing comprises providing the trained ANN to a printing device.
16. The computing device of claim 9 , wherein the training channel corresponds to a lightness channel, wherein the one or more ground truth channels correspond to one or more color channels, wherein the training involves predicting one or more output color channels from the lightness channel.
17. An article of manufacture comprising non-transitory data storage storing at least computer-readable instructions that, when executed by one or more processors of a computing device, cause the computing device to perform operations comprising:
receiving an input image comprising a plurality of channels;
determining a saliency map for the input image;
establishing at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels;
training an artificial neural network (ANN) to predict one or more output channels from the training channel, wherein the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and wherein the difference is computationally biased based on values from the saliency map; and
providing the trained ANN.
18. The article of manufacture of claim 17 , wherein the difference between the one or more output channels and the one or more ground truth channels comprises a pixel-wise difference between the one or more output channels and the one or more ground truth channels.
19. The article of manufacture of claim 18 , wherein the saliency map comprises respective values for each pixel in the input image, and wherein the computational bias involves scaling the pixel-wise difference between the one or more output channels and the one or more ground truth channels by the respective values.
20. The article of manufacture of claim 17 , wherein the input image is associated with at least one object class, and wherein the saliency map is determined with respect to the at least one object class.