image-resizing

This repository includes the source code, tooling and statistical analysis for the SE R&D research paper.

Application

Setup (macOS)

1. Install the C++ compiler through Xcode: xcode-select --install
2. Install the ImageMagick programs through Brew: brew install imagemagick
3. Install the just recipe runner through Brew: brew install just

Setup (GPUs)

1. Deploy a server with an NVIDIA GPU, e.g. from TensorDock
2. Install the CUDA toolkit from developer.nvidia.com
3. Install the ImageMagick programs through apt: sudo apt install imagemagick
4. Install the just recipe runner, see casey/just.

Usage

1. mkdir build && mkdir input_images
2. Generate the input images from a URL using just gen_inputs, e.g. just gen_inputs https://4kwallpapers.com/images/wallpapers/mclaren-w1-supercar-7680x4320-19246.jpg
3. Now you can run the application: run just run <platform> <algorithm>, e.g. just run cpu bilinear-interpolation. This will write the downsized image to build/output.rgba
4. To convert the image from RGBA to PNG and open it, run just render <size>, where <size> is the output resolution configured by the scaling factor in src/constants.hpp. Example: just render 1920x1080

Benchmarks

• Raw data was generated using benchmarking/runner/run.js, which runs each configuration (across factors, platforms and algorithms) 10 times.
• Data is aggregated to a CSV file using benchmarking/runner/aggregate.js.
• CSV file is read by the Jupyter notebooks, which use pandas and matplotlib to generate graphs. See bar, line and energy.