ARM-software / ComputeLibrary

The Compute Library is a set of computer vision and machine learning functions optimised for both Arm CPUs and GPUs using SIMD technologies.
MIT License
2.85k stars 776 forks source link

Conv2DBackpropInput layer in ACL? #795

Closed abhajaswal closed 4 years ago

abhajaswal commented 4 years ago

I am trying to check the support for Conv2DBackpropInput in arm compute library. I Checked the path ComputeLibrary-19.11\arm_compute\runtime\CL\functions for required interface , but could not locate it. Can you please let me know if this is supported in release v 19.11?

abhajaswal commented 4 years ago

Hi @morgolock can you please let me know about it!

abhajaswal commented 4 years ago

I want to implement ACL solution for UNET model for image segmentation - Model for same is at location https://github.com/PINTO0309/TensorflowLite-UNet/tree/master/model/semanticsegmentation_frozen_person_08.pb

This contains the Conv2DBackpropInput & Strided slice and pack operations.

Do i need to fold these operations by generating pb file using tflite to use deconvolution layer instead of these operations using ACL?

GeorgeARM commented 4 years ago

Hello @abhajaswal,

ComputeLibrary is designed for inference thus you won't find any training operations in place. In other words, we do not have a function to compute the gradients of a convolution. Hope this helps.