Business discovery from imagery
Aspects of the present disclosure relate to a method includes training a deep neural network using training images and data identifying one or more business storefront locations in the training images. The deep neural network outputs tight bounding boxes on each image. At the deep neural network, a first image may be received. The first image may be evaluated using the deep neural network. Bounding boxes may then be generated identifying business storefront locations in the first image.
1. A method comprising:
training, using one or more computing devices, a deep neural network using a set of training images and data identifying one or more business storefront locations in the training images, the deep neural network outputting a first plurality of bounding boxes on each training image;
receiving, using the one or more computing devices, a first image;
evaluating, using the one or more computing devices and the deep neural network, the first image; and
generating, using the one or more computing devices and the deep neural network, a second plurality of bounding boxes identifying two or more business storefront locations in the first image.
2. The method of claim 1 , further comprising:
detecting, using the one or more computing devices and the deep neural network, business information at each of the identified business storefront locations;
updating, using the one or more computing devices, a database of business information by adding information from each bounding box in the second plurality of bounding boxes with the business information detected at the business storefront location identified by the bounding box;
receiving, using the one or more computing devices, a request from a user for business information; and
retrieving, using the one or more computing devices, the requested business information from the updated database.
3. The method of claim 1 , wherein the second plurality of bounding boxes includes two bounding boxes arranged side by side in the first image identifying two discrete business storefront locations.
4. The method of claim 1 , wherein training the deep neural network further comprises:
applying a coarse sliding window on a portion of a given training image; and
removing one or more bounding boxes based on a location of the portion of the given training image.
5. The method of claim 1 , wherein generating the second plurality of bounding boxes further comprises:
applying a coarse sliding window on a portion of the first image; and
removing one or more bounding boxes based on a location of the portion of the given training image.
6. The method of claim 1 , wherein training the deep neural network further comprises:
determining a confidence score for each bounding box that represents a likelihood that the bounding box contains an image of a business storefront; and
removing bounding boxes corresponding to bounding boxes with a confidence score less than a set threshold.
7. The method of claim 1 , wherein generating the second plurality of bounding boxes further comprises:
determining confidence scores for each bounding box that represents a likelihood that the bounding box contains an image of a business storefront; and
removing bounding boxes locations corresponding to bounding boxes with a confidence score less than a set threshold.
8. The method of claim 1 , wherein:
training the deep neural network further comprises using post-classification; and
generating the second plurality of bounding boxes further comprises using post-classification.
9. The method of claim 1 , wherein generating the second plurality of bounding boxes further comprises:
calculating a probability of a given bounding box containing a business storefront;
ranking the second plurality of bounding boxes based on the calculated probability; and
removing one or more bounding boxes based on the ranking.
10. The method of claim 1 , wherein generating the second plurality of bounding boxes further comprises removing objects in the second plurality of bounding boxes that obstructs the view of the identified business storefront locations.
11. The method of claim 1 , wherein the training images and the first image are panoramic.
12. A system comprising:
a deep neural network; and
one or more computing devices configured to:
train the deep neural network using a set of training images and data identifying one or more business storefront locations in the training images, the deep neural network outputting a first plurality of bounding boxes on each training image;
receive, at the deep neural network, a first image;
evaluate, using the deep neural network, the first image; and
generate, using the deep neural network, a second plurality of bounding boxes identifying business storefront locations in the first image.
13. The system of claim 12 , wherein the one or more computing devices are further configured to train the deep neural network by:
applying a coarse sliding window on a portion of a given training image; and
removing one or more bounding boxes based on a location of the portion of the given training image.
14. The system of claim 12 , wherein the one or more computing devices are further configured to generate the second plurality of bounding boxes by:
applying a coarse sliding window on a portion of the first image; and
removing one or more bounding boxes based on a location of the portion of the given training image.
15. The system of claim 12 , wherein the one or more computing devices are further configured to train the deep neural network by:
determining a confidence score for each bounding box that represents a likelihood that the bounding box contains an image of a business storefront; and
removing bounding boxes corresponding to bounding boxes with a confidence score less than a set threshold.
16. The system of claim 12 , wherein the one or more computing devices are further configured to generate the second plurality of bounding boxes by:
determining confidence scores for each bounding box that represents a likelihood that the bounding box contains an image of a business storefront; and
removing bounding boxes locations corresponding to bounding boxes with a confidence score less than a set threshold.
17. The system of claim 12 , wherein the one or more computing devices are further configured to:
train the deep neural network by using post-classification; and
generate the second plurality of bounding boxes by using post-classification.
18. The system of claim 12 , wherein the one or more computing devices are further configured to generate the second plurality of bounding boxes by:
calculating a probability of a given bounding box containing a business storefront;
ranking the second plurality of bounding boxes based on the calculated probability; and
removing one or more bounding boxes based on the ranking.
19. The system of claim 12 , wherein the one or more computing devices are further configured to generate the second plurality of bounding boxes by removing objects in the second plurality of bounding boxes that obstructs the view of the identified business storefront locations.
20. A non-transitory, tangible computer-readable storage medium on which computer readable instructions of a program are stored, the instructions, when executed by one or more computing devices, cause the one or more computing devices to perform a method, the method comprising:
training a deep neural network using a set of training images and data identifying one or more business storefront locations in the training images, the deep neural network outputting a first plurality of bounding boxes on each training image;
receiving, at the deep neural network, a first image;
evaluating, using the deep neural network, the first image; and
generating, using the deep neural network, a second plurality of bounding boxes identifying business storefront locations in the first image.