Efficient construction and consumption of auxiliary channels in convolutional neural networks
An example device for processing image data includes a memory configured to store image data; and a processing system comprising one or more processors implemented in circuitry, the processing system being configured to: obtain an image to be processed; obtain a first auxiliary value for the image and a second auxiliary value for the image; generate an input component including the first auxiliary value and the second auxiliary value arranged in a pattern according to a stride of a neural network; and provide the image and the input component to the neural network.
1. A method of processing image data, the method comprising:
obtaining an image to be processed;
obtaining a first auxiliary value for the image and a second auxiliary value for the image;
generating an input component including the first auxiliary value and the second auxiliary value arranged in a pattern according to a stride of a neural network; and
providing the image and the input component to the neural network.
2. The method of claim 1 , wherein the image includes a first color component, a second color component, and a third color component, and wherein providing the image comprises providing the first color component, the second color component, and the third color component to the neural network.
3. The method of claim 2 , wherein the first color component comprises a red component, the second color component comprises a green component, and the third color component comprises a blue component.
4. The method of claim 2 , wherein the first color component comprises a luminance component, the second color component comprises a blue hue chrominance component, and the third color component comprises a red hue chrominance component.
5. The method of claim 1 , wherein the stride of the neural network represents a number of values of the image and the input component that the neural network processes per processing step.
6. The method of claim 1 , wherein the stride of the neural network is 2, and wherein generating the input component comprises generating the input component such that each sub-component of the input component, where the sub-components include at least 2 values, includes the first auxiliary value and the second auxiliary value.
7. The method of claim 1 , further comprising obtaining a third auxiliary value for the image and a fourth auxiliary value for the image, wherein generating the input component comprises generating the input component to include the third auxiliary value and the fourth auxiliary value.
8. The method of claim 7 , wherein generating the input component comprises:
generating each even-numbered row of the input component to include a repeating pattern of the first auxiliary value followed by the second auxiliary value; and
generating each odd-numbered row of the input component to include a repeating pattern of the third auxiliary value followed by the fourth auxiliary value.
9. The method of claim 7 , wherein the image was captured by a camera at a location at a time, the location having an X-component value, a Y-component value, and a Z-component value, wherein the camera had a yaw rotation value at the time, and wherein the first auxiliary value represents the X-component value, the second auxiliary value represents the Y-component value, the third auxiliary value represents the Z-component value, and the fourth auxiliary value represents the yaw rotation value.
10. The method of claim 1 , wherein the image has a number of samples, and wherein the input component has a number of samples equal to the number of samples of the image.
11. The method of claim 1 , wherein the image is obtained from a camera of a vehicle, the method further comprising:
receiving, from the neural network, data representing positions of objects relative to a position of the vehicle; and
at least partially autonomously controlling the vehicle according to the positions of the objects.
12. A device for processing image data, the device comprising:
a memory configured to store image data; and
a processing system comprising one or more processors implemented in circuitry, the processing system being configured to:
obtain an image to be processed;
obtain a first auxiliary value for the image and a second auxiliary value for the image;
generate an input component including the first auxiliary value and the second auxiliary value arranged in a pattern according to a stride of a neural network; and
provide the image and the input component to the neural network.
13. The device of claim 12 , wherein the image includes a first color component, a second color component, and a third color component, and wherein providing the image comprises providing the first color component, the second color component, and the third color component to the neural network.
14. The device of claim 13 , wherein the first color component comprises a red component, the second color component comprises a green component, and the third color component comprises a blue component.
15. The device of claim 13 , wherein the first color component comprises a luminance component, the second color component comprises a blue hue chrominance component, and the third color component comprises a red hue chrominance component.
16. The device of claim 12 , wherein the stride of the neural network represents a number of values of the image and the input component that the neural network processes per processing step.
17. The device of claim 12 , wherein the stride of the neural network is 2, and wherein generating the input component comprises generating the input component such that each sub-component of the input component, where the sub-components include at least 2 values, includes the first auxiliary value and the second auxiliary value.
18. The device of claim 12 , wherein the processing system is further configured to obtain a third auxiliary value for the image and a fourth auxiliary value for the image, wherein to generate the input component, the processing system is configured to generate the input component to include the third auxiliary value and the fourth auxiliary value.
19. The device of claim 18 , wherein to generate the input component, the processing system is configured to:
generate each even-numbered row of the input component to include a repeating pattern of the first auxiliary value followed by the second auxiliary value; and
generate each odd-numbered row of the input component to include a repeating pattern of the third auxiliary value followed by the fourth auxiliary value.
20. The device of claim 18 , wherein the image was captured by a camera at a location at a time, the location having an X-component value, a Y-component value, and a Z-component value, wherein the camera had a yaw rotation value at the time, and wherein the first auxiliary value represents the X-component value, the second auxiliary value represents the Y-component value, the third auxiliary value represents the Z-component value, and the fourth auxiliary value represents the yaw rotation value.
21. The device of claim 12 , wherein the image has a number of samples, and wherein the input component has a number of samples equal to the number of samples of the image.
22. The device of claim 12 , wherein the image is obtained from a camera of a vehicle, and wherein the processing system is further configured to:
receive, from the neural network, data representing positions of objects relative to a position of the vehicle; and
at least partially autonomously control the vehicle according to the positions of the objects.
23. A device for processing image data, the device comprising:
means for obtaining an image to be processed;
means for obtaining a first auxiliary value for the image and a second auxiliary value for the image;
means for generating an input component including the first auxiliary value and the second auxiliary value arranged in a pattern according to a stride of a neural network; and
means for providing the image and the input component to the neural network.
24. The device of claim 23 , wherein the image includes a first color component, a second color component, and a third color component, and wherein the means for providing the image comprises means for providing the first color component, the second color component, and the third color component to the neural network.
25. The device of claim 24 , wherein the first color component comprises a red component, the second color component comprises a green component, and the third color component comprises a blue component.
26. The device of claim 24 , wherein the first color component comprises a luminance component, the second color component comprises a blue hue chrominance component, and the third color component comprises a red hue chrominance component.
27. The device of claim 23 , wherein the stride of the neural network represents a number of values of the image and the input component that the neural network processes per processing step.
28. The device of claim 23 , wherein the stride of the neural network is 2, and wherein the means for generating the input component comprises means for generating the input component such that each sub-component of the input component, where the sub-components include at least 2 values, includes the first auxiliary value and the second auxiliary value.
29. The device of claim 23 , further comprising means for obtaining a third auxiliary value for the image and a fourth auxiliary value for the image, wherein the means for generating the input component comprises means for generating the input component to include the third auxiliary value and the fourth auxiliary value.
30. The device of claim 29 , wherein the means for generating the input component comprises:
means for generating each even-numbered row of the input component to include a repeating pattern of the first auxiliary value followed by the second auxiliary value; and
means for generating each odd-numbered row of the input component to include a repeating pattern of the third auxiliary value followed by the fourth auxiliary value.
31. The device of claim 29 , wherein the image was captured by a camera at a location at a time, the location having an X-component value, a Y-component value, and a Z-component value, wherein the camera had a yaw rotation value at the time, and wherein the first auxiliary value represents the X-component value, the second auxiliary value represents the Y-component value, the third auxiliary value represents the Z-component value, and the fourth auxiliary value represents the yaw rotation value.
32. The device of claim 23 , wherein the image has a number of samples, and wherein the input component has a number of samples equal to the number of samples of the image.
33. The device of claim 23 , wherein the image is obtained from a camera of a vehicle, further comprising:
means for receiving, from the neural network, data representing positions of objects relative to a position of the vehicle; and
means for at least partially autonomously controlling the vehicle according to the positions of the objects.