Processing images using deep neural networks

Patent metadata

Field	Value
Publication number	US-10977529-B2
Application number	US-202016846924-A
Country	US
Kind code	B2
Filing date	Apr 13, 2020
Priority date	Aug 29, 2014
Publication date	Apr 13, 2021
Grant date	Apr 13, 2021

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 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 implement:
a deep neural network configured to process data characterizing an input image to generate an alternative representation of the input image, the deep neural network comprising:
a plurality of subnetworks arranged in a sequence from lowest to highest, the plurality of subnetworks configured to process the data according to the sequence, the plurality of subnetworks comprising a plurality of module subnetworks, each of the module subnetworks comprising:
a pass-through convolutional layer configured to process a preceding output representation generated by a preceding subnetwork in the sequence and generate a pass-through output,
a plurality of groups of neural network layers configured to process the preceding output representation to generate a respective group output for each of the plurality of groups, wherein a first group of the plurality of groups includes at least two successive convolutional layers, a second group of the plurality of groups includes at least two successive convolutional layers, and a third group of the plurality of groups includes a pooling layer followed by a 1×1 convolutional layer, and
a concatenation layer configured to concatenate the pass-through output and the group outputs to generate an output representation for the module subnetwork; and

an output layer configured to process the alternative representation of the input image to generate an output from the input image.

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

3. The system of claim 1 , wherein the at least two successive convolutional layers included in the first group of the plurality of groups comprises a 1×1 convolutional layer followed by a 3×3 convolutional layer.

4. The system of claim 3 , wherein the at least two successive convolutional layers included in the second group of the plurality of groups comprises a 1×1 convolutional layer followed by a 3×3 convolutional layer.

5. The system of claim 1 , wherein the at least two successive convolutional layers included in the first group of the plurality of groups of neural network layers includes a 1×1 convolutional layer followed by a 1×7 convolutional layer.

6. The system of claim 5 , wherein the at least two successive convolutional layers included in the second group of the plurality of groups of neural network layers includes a 1×1 convolutional layer followed by a 1×7 convolutional layer.

7. The system of claim 1 , wherein the at least two successive convolutional layers included in the first group of the plurality of groups of neural network layers includes a 1×1 convolutional layer followed by a 1×3 convolutional layer.

8. The system of claim 7 , wherein the at least two successive convolutional layers included in the second group of the plurality of groups of neural network layers includes a 1×1 convolutional layer followed by a 1×3 convolutional layer.

9. The system of claim 1 , wherein the output layer is a softmax output layer.

10. The system of claim 1 , wherein the plurality of subnetworks comprises a training subnetwork that includes one or more average pooling layers.

11. The system of claim 1 , wherein the plurality of subnetworks comprise an additional module subnetwork, the additional module subnetwork comprising:
a second plurality of groups of neural network layers configured to process the preceding output representation to generate a respective group output for each of second plurality of groups, wherein a first group of the second plurality of groups includes a 1×1 convolutional layer followed by a 3×3 convolutional layer, a second group of the second plurality of groups includes a 1×1 convolutional layer followed by a 1×7 convolutional layer, and a third group of the second plurality of groups includes a max-pooling layer; and
a concatenation layer configured to concatenate the outputs of the second plurality of groups of neural network layers to generate an output representation for the additional module subnetwork.

12. The system of claim 11 , wherein the max-pooling layer is a 3×3 max-pooling layer.

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

14. The system of claim 1 , wherein the alternative representation of the input image is an image classification of the input image.

15. The system of claim 1 , wherein the alternative representation of the input image includes data for one or more objects detected in the input image, including a location of the one or more objects in the input image or a size of the one or more objects in the input image.

16. The system of claim 1 , wherein the plurality of subnetworks comprise an additional module subnetwork, the additional module subnetwork comprising:
a second plurality of groups of neural network layers configured to process the preceding output representation to generate a respective group output for each of second plurality of groups, wherein a first group of the second plurality of groups includes a 1×1 convolutional layer followed by a 3×3 convolutional layer, a second group of the second plurality of groups includes a 3×3 convolutional layer, and a third group of the second plurality of groups includes a 3×3 max-pooling layer; and
a concatenation layer configured to concatenate the outputs of the second plurality of groups to generate an output representation for the additional module subnetwork.

Assignees

• Google Llc

Inventors

• Szegedy Christian
• Vanhoucke Vincent O

Classifications

• G06N3/084 Primary

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

• G06N3/045 Primary

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• H04N19/00

  Methods or arrangements for coding, decoding, compressing or decompressing digital video signals · CPC title