Image classification through label progression
Systems and methods are disclosed for training neural networks using labels for training data that are dynamically refined using neural networks and using these trained neural networks to perform detection and/or classification of one or more objects appearing in an image. Particular embodiments may generate a set of crops of images from a corpus of images, then apply a first neural network to the set of crops to obtain a set of respective outputs. A second neural network may then be trained using the set of crops as training examples. The set of respective outputs may be applied as labels for the set of crops.
1. A method comprising:
obtaining a first image from an image sensor;
applying, by a computing device to the first image, a neural network that has been trained using training data that includes a set of crops of an image and a set of respective outputs from another neural network that are applied as labels for the set of crops;
modifying, by the computing device, at least one parameter for operating the image sensor based on an output of the neural network; and
obtaining, by the computing device, a second image with the image sensor using the at least one modified parameter.
2. The method of claim 1 , wherein the image sensor is an image sensor of the computing device.
3. The method of claim 1 , wherein obtaining the first image comprises obtaining the first image at the computing device over a communications link from a remote image capture system that includes the image sensor.
4. The method of claim 1 , wherein the output of the neural network comprises classification data for object in the first image.
5. The method of claim 1 , wherein the at least one parameter comprises an electronic shutter setting.
6. The method of claim 1 , wherein the at least one parameter comprises at least one of: an exposure time, shutter speed, a high dynamic range setting, and an electronic rolling shutter setting.
7. The method of claim 1 , wherein the output of the neural network comprises a class of a scene in the first image.
8. The method of claim 7 , wherein modifying the at least one parameter of the image sensor comprises setting the at least one parameter to a value associated with the class of the scene.
9. A system, comprising:
a processing apparatus configured to:
obtain a first image from an image sensor;
apply, to the first image, a neural network that has been trained using training data that includes a set of crops of an image and a set of respective outputs from another neural network that are applied as labels for the set of crops;
modify at least one parameter for operating the image sensor based on an output of the neural network; and
obtain a second image with the image sensor using the at least one modified parameter.
10. The system of claim 9 , wherein the at least one parameter comprises at least one of: an exposure time, shutter speed, a high dynamic range setting, and an electronic rolling shutter setting.
11. The system of claim 9 , further comprising the image sensor.
12. The system of claim 9 , wherein output of the neural network comprises a class of an object or a scene in the first image.
13. A non-transitory machine-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
obtaining, by a computing device, a first image from an image sensor;
applying, by the computing device to the first image, a neural network that has been trained using training data that includes a set of crops of an image and a set of respective outputs from another neural network that are applied as labels for the set of crops;
modifying, by the computing device, at least one parameter for operating the image sensor based on an output of the neural network; and
obtaining, by the computing device, a second image with the image sensor using the at least one modified parameter.
14. The non-transitory machine-readable medium of claim 13 , wherein the image sensor is an image sensor of the computing device.
15. The non-transitory machine-readable medium of claim 13 , wherein obtaining the first image comprises obtaining the first image at the computing device over a communications link from a remote image capture system that includes the image sensor.
16. The non-transitory machine-readable medium of claim 13 , wherein the output of the neural network comprises classification data for object in the first image.
17. The non-transitory machine-readable medium of claim 13 , wherein the at least one parameter comprises an electronic shutter setting.
18. The non-transitory machine-readable medium of claim 13 , wherein the at least one parameter comprises at least one of: an exposure time, shutter speed, a high dynamic range setting, and an electronic rolling shutter setting.
19. The non-transitory machine-readable medium of claim 13 , wherein the output of the neural network comprises a class of a scene in the first image.
20. The non-transitory machine-readable medium of claim 19 , wherein modifying the at least one parameter of the image sensor comprises setting the at least one parameter to a value associated with the class of the scene.