Generating object proposals using deep-learning models
View Patent ↗In one embodiment, a plurality of patches of an image are processed using a first deep-learning model to detect a plurality of features associated with the first patch of the image. Each patch includes one or more pixels of the image. Using a second deep-learning model, a respective object proposal is generated for each of the plurality of patches of the image. The second deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and each object proposal includes a prediction as to a location of an object in the patch. Using a third deep-learning model, a respective score is computed for each object proposal generated using the second deep-learning model. The third deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and the object score may include a likelihood that the patch contains an entire object.
1. A method comprising:
processing, by one or more computing devices, a plurality of patches of an image, using a first deep-learning model, to detect a plurality of features associated with each patch of the image, wherein each patch comprises one or more pixels of the image;
generating, by one or more computing devices, using a second deep-learning model, a respective object proposal for each of the plurality of patches of the image, wherein the second deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein each object proposal comprises a prediction as to a location of an object in the respective patch; and
computing, by one or more computing devices, using a third deep-learning model, a respective score for each object proposal generated using the second deep-learning model, wherein the third deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein the score comprises a likelihood that the patch contains an entire object.
2. The method of claim 1 , wherein the first deep-learning model, the second deep-learning model, and the third deep-learning model comprise convolutional neural networks.
3. The method of claim 1 , further comprising:
ranking the plurality of object proposals based on the respective scores; and
determining a subset of object proposals of the plurality of object proposals based on the ranking.
4. The method of claim 3 , further comprising generating identifications of one of more objects in the image based on the subset of object proposals.
5. The method of claim 1 , wherein one or more of the plurality of patches are overlapping.
6. The method of claim 1 , further comprising determining the plurality of patches using a window of a fixed size.
7. The method of claim 6 , wherein the fixed size of the window is 200 pixels by 200 pixels.
8. The method of claim 6 , wherein the window is shifted by 16 pixels to determine each of the plurality of patches.
9. The method of claim 6 , further comprising determining the plurality of patches by scaling the image.
10. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
process a plurality of patches of an image, using a first deep-learning model, to detect a plurality of features associated with each patch of the image, wherein each patch comprises one or more pixels of the image;
generate, using a second deep-learning model, a respective object proposal for each of the plurality of patches of the image, wherein the second deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein each object proposal comprises a prediction as to a location of an object in the respective patch; and
compute, using a third deep-learning model, a respective score for each object proposal generated using the second deep-learning model, wherein the third deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein the score comprises a likelihood that the patch contains an entire object.
11. The media of claim 10 , wherein the first deep-learning model, the second deep-learning model, and the third deep-learning model comprise convolutional neural networks.
12. The media of claim 10 , further comprising:
ranking the plurality of object proposals based on the respective scores; and
determining a subset of object proposals of the plurality of object proposals based on the ranking.
13. The media of claim 12 , further comprising generating identifications of one of more objects in the image based on the subset of object proposals.
14. The media of claim 10 , wherein one or more of the plurality of patches are overlapping.
15. A system comprising: one or more processors; and a memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
process a plurality of patches of an image, using a first deep-learning model, to detect a plurality of features associated with each patch of the image, wherein each patch comprises one or more pixels of the image;
generate, using a second deep-learning model, a respective object proposal for each of the plurality of patches of the image, wherein the second deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein each object proposal comprises a prediction as to a location of an object in the respective patch; and
compute, using a third deep-learning model, a respective score for each object proposal generated using the second deep-learning model, wherein the third deep-learning model takes as input the plurality of detected features associated with the respective patch of the image, and wherein the score comprises a likelihood that the patch contains an entire object.
16. The system of claim 15 , wherein the first deep-learning model, the second deep-learning model, and the third deep-learning model comprise convolutional neural networks.
17. The system of claim 15 , further comprising:
ranking the plurality of object proposals based on the respective scores; and
determining a subset of object proposals of the plurality of object proposals based on the ranking.
18. The system of claim 17 , further comprising generating identifications of one of more objects in the image based on the subset of object proposals.
19. The system of claim 15 , wherein one or more of the plurality of patches are overlapping.