Remote distribution of neural networks
Remote distribution of multiple neural network models to various client devices over a network can be implemented by identifying a native neural network and remotely converting the native neural network to a target neural network based on a given client device operating environment. The native neural network can be configured for execution using efficient parameters, and the target neural network can use less efficient but more precise parameters.
1. A method comprising:
detecting, from a device, a triggering event of a neural network conversion process associated with an image effect;
identifying, by one or more hardware processors, a native neural network model associated with the image effect;
converting the native neural network model to a target neural network model based on computational resources associated with the device;
in response to converting the native neural network model to a target native neural network model, causing display of user interface element associated with the image effect on a user interface of the device; and
generating a modified image by applying the target neural network model to an image.
2. The method of claim 1 , wherein the triggering event of the neural network conversion process comprises an application event that corresponds to any one of: navigation to a pre-specified user interface or activation of a messaging client application associated with the device.
3. The method of claim 1 , comprising:
determining the target neural network model based on an operating system associated with the device having libraries for parallel execution of neural networks on a plurality of hardware processors.
4. The method of claim 3 , wherein the target neural network model is associated with input parameters that are not included in the native neural network model.
5. The method of claim 1 , comprising:
identifying computational hardware of the device; and
modifying a size of the image to an input image size based on an identified computational hardware of the device.
6. The method of claim 1 , comprising:
specifying a plurality of input parameters of the target neural network model; and
compiling the target neural network model using the plurality of input parameters.
7. The method of claim 6 , wherein the plurality of input parameters includes a first input parameter corresponding to an input image size of the image.
8. The method of claim 1 , wherein the computational resources associated with the device comprises one or more of: a screen size, bandwidth quality, GPU availability, one or more hardware acceleration libraries, multi-core parallel processing capability on a CPU, and multi-core concurrent processing capability on the CPU.
9. The method of claim 1 , comprising:
receiving usage data from the device; and
identifying the native neural network model associated with the image effect based on the usage data.
10. The method of claim 9 , wherein the usage data comprises frequency statistics of neural network model conversion.
11. A system comprising:
a memory storing instructions; and
one or more hardware processors communicatively coupled to the memory and configured by the instructions to perform operations comprising:
detecting, from a device, a triggering event of a neural network conversion process associated with an image effect;
identifying a native neural network model associated with the image effect;
converting the native neural network model to a target neural network model based on computational resources associated with the device;
in response to converting the native neural network model to a target native neural network model, causing display of user interface element associated with the image effect on a user interface of the device; and
generating a modified image by applying the target neural network model to an image.
12. The system of claim 11 , wherein the triggering event of the neural network conversion process comprises an application event that corresponds to any one of: navigation to a pre-specified user interface or activation of a messaging client application associated with the device.
13. The system of claim 11 , wherein the operations comprise:
determining the target neural network model based on an operating system associated with the device having libraries for parallel execution of neural networks on a plurality of hardware processors.
14. The system of claim 13 , wherein the target neural network model is associated with input parameters that are not included in the native neural network model.
15. The system of claim 11 , wherein the operations comprise:
identifying computational hardware of the device; and
modifying a size of the image to an input image size based on an identified computational hardware of the device.
16. The system of claim 11 , wherein the operations comprise:
specifying a plurality of input parameters of the target neural network model; and
compiling the target neural network model using the plurality of input parameters.
17. The system of claim 16 , wherein the plurality of input parameters includes a first input parameter corresponding to an input image size of the image.
18. The system of claim 11 , wherein the computational resources associated with the device comprises one or more of: a screen size, bandwidth quality, GPU availability, one or more hardware acceleration libraries, multi-core parallel processing capability on a CPU, and multi-core concurrent processing capability on the CPU.
19. The system of claim 11 , wherein the operations comprise:
receiving usage data from the device; and
identifying the native neural network model associated with the image effect based on the usage data.
20. A non-transitory machine-readable storage medium comprising instructions that, when executed by a processing device, cause the processing device to perform operations comprising:
detecting, from a device, a triggering event of a neural network conversion process associated with an image effect;
identifying a native neural network model associated with the image effect;
converting the native neural network model to a target neural network model based on computational resources associated with the device;
in response to converting the native neural network model to a target native neural network model, causing display of user interface element associated with the image effect on a user interface of the device; and
generating a modified image by applying the target neural network model to an image.