Processing images using deep neural networks

Patent metadata

Field	Value
Publication number	US-9715642-B2
Application number	US-201514839452-A
Country	US
Kind code	B2
Filing date	Aug 28, 2015
Priority date	Aug 29, 2014
Publication date	Jul 25, 2017
Grant date	Jul 25, 2017

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for image processing using deep neural networks. One of the methods includes receiving data characterizing an input image; processing the data characterizing the input image using a deep neural network to generate an alternative representation of the input image, wherein the deep neural network comprises a plurality of subnetworks, wherein the subnetworks are arranged in a sequence from lowest to highest, and wherein processing the data characterizing the input image using the deep neural network comprises processing the data through each of the subnetworks in the sequence; and processing the alternative representation of the input image through an output layer to generate an output from the input image.

First claim

What is claimed is:

1. A method comprising:
receiving data characterizing an input image;
processing the data characterizing the input image using a deep neural network to generate an alternative representation of the input image, wherein the deep neural network comprises a plurality of subnetworks, wherein the subnetworks are arranged in a sequence from lowest to highest, and wherein processing the data characterizing the input image using the deep neural network comprises processing the data through each of the subnetworks in the sequence, wherein the plurality of subnetworks comprise a plurality of module subnetworks, and wherein each of the module subnetworks is configured to:
receive a preceding output representation generated by a preceding subnetwork in the sequence;
process the preceding output representation through a pass-through convolutional layer to generate a pass-through output;
process the preceding output representation through one or more groups of neural network layers to generate a respective group output for each of the one or more groups; and
concatenate the pass-through output and the group outputs to generate an output representation for the module subnetwork; and

processing the alternative representation of the input image through an output layer to generate an output from the input image.

2. The method of claim 1 , wherein the pass-through convolutional layer is a 1×1 convolutional layer.

3. The method of claim 1 , wherein processing the preceding output representation through each of the one or more groups comprises:
processing the preceding output through each layer of a first group of neural network layers to generate a first group output, wherein the first group comprises a first convolutional layer followed by a second convolutional layer.

4. The method of claim 3 , wherein the first convolutional layer is a 1×1 convolutional layer.

5. The method of claim 3 , wherein the second convolutional layer is a 3×3 convolutional layer.

6. The method of claim 1 , wherein processing the preceding output using each of the one or more groups comprises:
processing the preceding output through each layer of a second group of neural network layers to generate a second group output, wherein the second group comprises a third convolutional layer followed by a fourth convolutional layer.

7. The method of claim 6 , wherein the third convolutional layer is a 1×1 convolutional layer.

8. The method of claim 6 , wherein the fourth convolutional layer is a 5×5 convolutional layer.

9. The method of claim 1 , wherein processing the preceding output using each of the one or more groups comprises:
processing the preceding output through each layer of a third group of neural network layers to generate a third group output, wherein the third group comprises a first max-pooling layer followed by a fifth convolutional layer.

10. The method of claim 9 , wherein the first max-pooling layer is a 3×3 max pooling layer.

11. The method of claim 9 , wherein the fifth convolutional layer is a 1×1 convolutional layer.

12. The method of claim 1 , wherein the plurality of subnetworks comprises one or more additional max-pooling layers.

13. The method of claim 1 , wherein the plurality of subnetworks comprises one or more initial convolutional layers.

14. A system comprising one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to perform operations comprising:
receiving data characterizing an input image;
processing the data characterizing the input image using a deep neural network to generate an alternative representation of the input image, wherein the deep neural network comprises a plurality of subnetworks, wherein the subnetworks are arranged in a sequence from lowest to highest, and wherein processing the data characterizing the input image using the deep neural network comprises processing the data through each of the subnetworks in the sequence, wherein the plurality of subnetworks comprise a plurality of module subnetworks, and wherein each of the module subnetworks is configured to:
receive a preceding output representation generated by a preceding subnetwork in the sequence;
process the preceding output representation through a pass-through convolutional layer to generate a pass-through output;
process the preceding output representation through one or more groups of neural network layers to generate a respective group output for each of the one or more groups; and concatenate the pass-through output and the group outputs to generate an output representation for the module subnetwork; and

processing the alternative representation of the input image through an output layer to generate an output from the input image.

15. The system of claim 14 , wherein processing the preceding output representation through each of the one or more groups comprises:
processing the preceding output through each layer of a first group of neural network layers to generate a first group output, wherein the first group comprises a first convolutional layer followed by a second convolutional layer.

16. The system of claim 14 , wherein processing the preceding output using each of the one or more groups comprises:
processing the preceding output through each layer of a second group of neural network layers to generate a second group output, wherein the second group comprises a third convolutional layer followed by a fourth convolutional layer.

17. The system of claim 14 , wherein processing the preceding output using each of the one or more groups comprises:
processing the preceding output through each layer of a third group of neural network layers to generate a third group output, wherein the third group comprises a first max-pooling layer followed by a fifth convolutional layer.

18. A computer program product encoded on one or more non-transitory computer storage media, the computer program product comprising instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
receiving data characterizing an input image;
processing the data characterizing the input image using a deep neural network to generate an alternative representation of the input image, wherein the deep neural network comprises a plurality of subnetworks, wherein the subnetworks are arranged in a sequence from lowest to highest, and wherein processing the data characterizing the input image using the deep neural network comprises processing the data through each of the subnetworks in the sequence, wherein the plurality of subnetworks comprise a plurality of module subnetworks, and wherein each of the module subnetworks is configured to:
receive a preceding output representation generated by a preceding subnetwork in the sequence;
process the preceding output representation through a pass-through convolutional layer to generate a pass-through output;
process the preceding output representation through one or more groups of neural network layers to generate a respective group output for each of the one or more groups; and concatenate the pass-through output and the group outputs to generate an output representation for the module subnetwork; and

processing the alternative representation of the input image through an output layer to generate an output from the input image.

19. The computer program product of claim 18 , wherein the pass-through convolutional layer is a 1×1 convolutional layer.

Assignees

• Google Inc

Inventors

• Szegedy Christian
• Vanhoucke Vincent O

Classifications

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06N3/063

  using electronic means · CPC title

• G06N3/045 Primary

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title