Systems and methods for image classification using a neural network combined with a correlation structure
Systems and methods for performing image classification are disclosed. The methods include, by a processor: receiving an input image, generating a label prediction corresponding to the input image using a trained neural network, generating a correlation structure based on a comparison of the input image with each of a plurality of reference images, and generating an updated label prediction corresponding to the input image using the label prediction and the correlation structure.
1 . A method comprising, by a processor:
receiving an input image;
generating, using a trained neural network, a label prediction corresponding to the input image;
generating, based on a comparison of the input image with each of a plurality of reference images, a correlation structure;
generating, using the label prediction and the correlation structure, an updated label prediction corresponding to the input image;
using the updated label prediction for controlling a navigation route of an autonomous vehicle; and
performing an autonomous driving action to the autonomous vehicle using the navigation route, wherein the autonomous driving action includes at least one of acceleration, braking, and steering.
2 . The method of claim 1 , wherein generating the correlation structure comprises computing a distance between the input image and each of the plurality of reference images, the distance being associated with one or more color channels.
3 . The method of claim 1 , wherein generating the updated label prediction corresponding to the input image comprises identifying the label prediction as the updated label prediction in response to determining that there exists a correlation between the input image and each of the plurality of reference images that is less than a threshold.
4 . The method of claim 1 , wherein generating the updated label prediction corresponding to the input image comprises identifying a correct label associated with a reference image as the updated label prediction in response to determining that there exists a high correlation between the input image and the reference image that outweighs the label prediction.
5 . The method of claim 1 , further comprising generating a reference dataset comprising the plurality of reference images, wherein at least one of the plurality of reference images comprises an adversarial image or an image associated with an edge case.
6 . The method of claim 5 , further comprising generating the reference dataset to include a correct label prediction associated with each of the plurality of reference images.
7 . The method of claim 1 , wherein generating the correlation structure comprises computing a distance between the input image and each of the plurality of reference images by comparing a feature map of the input image and a reference feature map of that reference image, the feature map and the reference feature map obtained from a layer of the trained neural network.
8 . The method of claim 1 , wherein generating the correlation structure comprises:
identifying a reference image of the plurality of reference images that has a highest correlation with the input image; and
identifying a correct label associated with the identified reference image as the updated label prediction.
9 . The method of claim 1 , wherein the correlation structure is a Gaussian process correlation structure.
10 . A system for performing image classification, the system comprising:
a processor; and
a non-transitory computer-readable medium comprising programming instructions that when executed by the processor will cause the processor to:
receive an input image,
generate, using a trained neural network, a label prediction corresponding to the input image,
generate, based on a comparison of the input image with each of a plurality of reference images, a correlation structure,
generate, using the label prediction and the correlation structure, an updated label prediction corresponding to the input image, and
use the updated label prediction for controlling navigation of an autonomous vehicle including at least one of: acceleration, braking, and steering.
11 . The system of claim 10 , wherein the programming instructions that when executed by the processor will cause the processor to generate the correlation structure comprise programming instructions to cause the processor to compute a distance between the input image and each of the plurality of reference images, the distance being associated with one or more color channels.
12 . The system of claim 10 , wherein the programming instructions that when executed by the processor will cause the processor to generate the updated label prediction corresponding to the input image comprise programming instructions to cause the processor to identify the label prediction as the updated label prediction in response to determining that there exists a correlation between the input image and each of the plurality of reference images that is less than a threshold.
13 . The system of claim 10 , wherein the programming instructions that when executed by the processor will cause the processor to generate the updated label prediction corresponding to the input image comprise programming instructions to cause the processor to identify a correct label associated with a reference image as the updated label prediction in response to determining that there exists a high correlation between the input image and the reference image that outweighs the label prediction.
14 . The system of claim 10 , further comprising programming instructions that when executed by the processor will cause the processor to generate a reference dataset comprising the plurality of reference images, wherein at least one of the plurality of reference images comprises an adversarial image or an image associated with an edge case.
15 . The system of claim 14 , further comprising programming instructions that when executed by the processor will cause the processor to generate the reference dataset to include a correct label prediction associated with each of the plurality of reference images.
16 . The system of claim 10 , wherein the programming instructions that when executed by the processor will cause the processor to generate the correlation structure comprise programming instructions to cause the processor to compute a distance between the input image and each of the plurality of reference images by comparing a feature map of the input image and a reference feature map of that reference image, the feature map and the reference feature map obtained from a layer of the trained neural network.
17 . The system of claim 10 , wherein the programming instructions that when executed by the processor will cause the processor to generate the correlation structure comprise programming instructions to cause the processor to:
identify a reference image of the plurality of reference images that has a highest correlation with the input image; and
identify a correct label associated with the identified reference image as the updated label prediction.
18 . A method for an autonomous vehicle, comprising:
receiving an input image;
generating, using a trained neural network, a label prediction corresponding to the input image;
generating, based on a comparison of the input image with each of a plurality of reference images, a correlation structure;
generating, using the label prediction and the correlation structure, an updated label prediction corresponding to the input image; and
controlling navigation of an autonomous vehicle using the updated label prediction, wherein the controlling navigation includes at least one of: acceleration, braking, and steering.
19 . An autonomous vehicle comprising:
a vehicle system for controlling at least one of acceleration, braking and steering; and
a system for performing image classification according to claim 10 .