CuImages

This is basically a playground for developing the GPU algorithms and support differentiable programming for image processing in Julia. The goal is to provide a set of GPU algorithms for image processing in Julia. The algorithms are implemented using CUDA.jl, CUDAKernels.jl and KernelAbstractions.jl

We want to currently focus on following topics:

• Augmentation
• Enchancement(adjust gamma, histogramequalisation, etc etc)
• Image Filtering

Installation:

julia> add https://github.com/ashwani-rathee/CuImages.jl.git

julia> using CuImages

Usage:

Focus of this package currently is on following operations:

• Convert Image to CuArray
• Apply Algorithm on it
• Convert CuArray to Image

Point of this current package is to learn:

• how to use CUDA.jl
• how to use CUDAKernels.jl
• how to use KernelAbstractions.jl
• Designing GPU algorithms, GPU kernels, CPU fallbacks and Parallel GPU computing
• Benchmark and learn what's fast
• Learn about the design choices that need to be made for GPU support and differentiable computer vision support

Now, how to convert AbstractArray to CuArray and vice versa:

julia> using CuImages, TestImages

julia> img = testimage("cameraman");

julia> img_gpu = Cu(img);

julia> img_cpu = Ar(img_gpu);

Now, how to apply algorithm on it:

julia> img_gpu = Cu(img);

julia> img_res = scale(img_gpu, 0.5, 0.2);

julia> img_res_cpu = Ar(img_res);

Using CuImages.jl, you can apply following algorithms on image.

For Augmentation, below mentioned algorithms are available:

• Scale Operation
• Rotate Operation
• Translate Operation

For Filtering, below mentioned algorithms are available:

• Gaussian Filter
• Sobel Filter
• Prewitt Filter
• Ando3 Filter
• Ando5 Filter
• Scharr Filter
• Bickley Filter

Contributions and Issues:

If you have questions about GIFImages.jl, feel free to get in touch via Slack or open an issue hearts