Modulated image segmentation
A modulated segmentation system can use a modulator network to emphasize spatial prior data of an object to track the object across multiple images. The modulated segmentation system can use a segmentation network that receives spatial prior data as intermediate data that improves segmentation accuracy. The segmentation network can further receive visual guide information from a visual guide network to increase tracking accuracy via segmentation.
1. A method comprising:
generating, by one or more processors of a user device, an image depicting an object;
generating shape parameters that describe a shape of the object;
generating spatial parameters that emphasize a location of the object depicted in the image;
generating image mask data for the object based on the shape parameters and the spatial parameters; and
generating a modified image from the image mask data and the image.
2. The method of claim 1 , wherein generating spatial parameters further comprises:
generating spatial parameters using a previous image that is generated prior to the image.
3. The method of claim 1 , further comprising:
generating multiple layer parameters using a first neural network, each layer parameter configured to modify intermediate feature data of a second neural network by:
generating the shape parameters that describe the shape of the object; and
generating the spatial parameters that emphasize the location of the object depicted in the image.
4. The method of claim 3 , wherein generating the image mask data further comprises:
generating the image mask data for the object using the second neural network, the second neural network comprising a plurality of intermediate layers configured to generate modulated feature data using the multiple layer parameters generated by the first neural network.
5. The method of claim 3 , further comprising:
training the first neural network and second neural network on training data using gradient descent.
6. The method of claim 3 , wherein the second neural network is trained on training data that does not include the object depicted in the image, and wherein the first neural network and the second neural network are trained using end-to-end training.
7. The method of claim 6 , wherein the training data comprises images of different objects.
8. The method of claim 7 , wherein the image mask data indicates pixel locations of one of the different objects.
9. The method of claim 3 , wherein the first neural network generates the multiple sets of layer parameters using, as inputs, a shape object and a spatial prior of the shape object, wherein the shape object has a different shape than a shape used to generate the image mask data, wherein the spatial prior is a Gaussian distribution.
10. The method of claim 1 , further comprising:
storing the image mask data on the user device; and
publishing the modified image as an ephemeral message on a network site.
11. A system comprising:
one or more processors of a machine; and
a memory storing instructions that, when executed by the one or more processors, cause the machine to perform operations comprising:
generating an image depicting an object;
generating shape parameters that describe a shape of the object;
generating spatial parameters that emphasize a location of the object depicted in the image;
generating image mask data for the object based on the shape parameters and the spatial parameters; and
generating a modified image from the image mask data and the image.
12. The system of claim 11 , wherein generating spatial parameters further comprises:
generating spatial parameters using a previous image that is generated prior to the image.
13. The system of claim 11 , wherein the operations further comprise:
generating multiple layer parameters using a first neural network, each layer parameter configured to modify intermediate feature data of a second neural network by:
generating the shape parameters that describe the shape of the object; and
generating the spatial parameters that emphasize the location of the object depicted in the image.
14. The system of claim 13 , wherein generating the image mask data further comprises:
generating the image mask data for the object using the second neural network, the second neural network comprising a plurality of intermediate layers configured to generate modulated feature data using the multiple layer parameters generated by the first neural network.
15. The system of claim 13 , wherein the operations further comprise:
training the first neural network and second neural network on training data using gradient descent.
16. The system of claim 13 , wherein the second neural network is trained on training data that does not include the object depicted in the image, and wherein the first neural network and the second neural network are trained using end-to-end training.
17. The system of claim 16 , wherein the training data comprises images of different objects.
18. The system of claim 17 , wherein the image mask data indicates pixel locations of one of the different objects.
19. The system of claim 13 , wherein the first neural network generates the multiple sets of layer parameters using, as inputs, a shape object and a spatial prior of the shape object, wherein the shape object has a different shape than a shape used to generate the image mask data, wherein the spatial prior is a Gaussian distribution.
20. A non-transitory machine-readable storage device embodying instructions that, when executed by a machine, cause the machine to perform operations comprising:
generating, by one or more processors of the machine, an image depicting an object;
generating shape parameters that describe a shape of the object;
generating spatial parameters that emphasize a location of the object depicted in the image;
generating image mask data for the object based on the shape parameters and the spatial parameters; and
generating a modified image from the image mask data and the image.