Neural Denoising + Python Examples
acle_ops: Renaming acle library to acle_ops. Operators are now grouped into separate namespaces describing the nature of the operator (for eg. filter, resize, elementwise etc.). Removing 'impl_' prefix from all functions and creating new namespace called dispatch in each operator to dynamically resolve functions when multiple kernels are available (i.e., neon, sve, sme2). A user can call both the dynamic dispatch function or choose a specific function to call in the appropriate namspace. All headers are now public and can be found in cpp/common/acle/include. Common headers are tidied up into sve.h, utils.h, multithread.h, and tile.h. Moving all pipeline specific code for collapsenet and ultralite to respective source files as this library is meant for generic re-usable operators only. Updating legacy multithread loop offset function to new one.
Neural Denoising: Pipeline specific implementations added to usecase/neural_denoiser/core and benchmarking wrapper logic in usecase/neural_denoiser/benchmark. Compiling 6 executables (spatial/temporal/spatio-temporal mode and 4K/1080p). Adding shell script to run benchmarking executables on custom inputs and weights loaded from disk by implementing some minimal pre/post processing steps in python. Adding Ultralite and CollapseNet checkpoints (not up to date - April revision) and 1 lux lab video sequence as test resources.
Cinematic Mode and Low Light Image Enhancement: Removed PPM library and modified cinematic mode and low light enhancement executables to use opencv imread/imwrite functions. Upgraded benchmark versions to most up-to-date current versions for both cinematic mode and low light enhancement. Copying fixes to HDRNetLIME code to work with float32 tflite for better image quality due to less post-training quantisation error. Updating image and tflite resources.
Python: Adding python requirements for virtual environment and self-contained implementations of the example pipelines in Keras and PyTorch to run on custom inputs.
Updating ReadMe. Also updating CI pipeline to reflect these changes.