nerovalerius/deep_learning_compression

deep_learning_compression

Folder with additional information: GoogleDrive

Howto

The main part of this project is the evaluation of the compression methods. The evaluation is located in a jupyter notebook in ./Notebooks/Evaluation.ipynb. To generate the data needed for this you have to run some scripts in the following order:

$ python ./Scripts/PrepareData.py
$ python ./Scripts/ResizeImagesInDataset.py
$ python ./Scripts/RunCompression.py
$ python ./Scripts/RunMetricEvaluation.py

The creates a file ./tmp/last_run_metrics.csv that needs to be imported into the evaluation jupyter notebook.

Installation

One should consider the usage of an virtual environment. Please see Creating virtual environments for more information.

To install the dependencies run:

$ python -m pip install -r requirements.txt

$ git submodules init && git submodule update

For Development purposes, one should also install the project locally with:

$ python -m pip install -e .

The following dependencies are required to use Compression.py:

• nconvert - download, build and install from https://www.xnview.com/en/xnconvert/#downloads
• openjpeg - download an build and install from https://github.com/uclouvain/openjpeg/blob/master/INSTALL.md

Datasets

To evaluate performance of the compression techniques, different datasets can be used. They will be located in the ./datasets folder.

Available Datasets:

• ImageCompressionBenchmark8BitRGB

To make the data available locally, a script was created. This downloads all different datasets.

$ python Scripts/PrepareData.py

Found following datasets:
	- ImageCompressionBenchmark8BitRGB


INFO:root:Found total 1 datasets that will be downloaded.
INFO:root:Dataset ImageCompressionBenchmark8BitRGB was already downloaded.
INFO:root:Dataset ImageCompressionBenchmark8BitRGB was already prepared!

To use the data of a specific dataset, you can use the storage interface:

from deep_learning_compression.Storage.Storage import Storage

s = Storage()  # Make an instance

s.GetAllFileInformationOfASingleDataSet(setName="ImageCompressionBenchmark8BitRGB")

Compression

To run the compression over all methods, you can run following script. (Currently, the values are hard coded, but this can easily be extended to make variable options!)

$ python ./Scripts/RunCompression.py

Image Quality Metrics

We use full- and no-reference image quality metrics to assess the quality of the used compression techniques.

Full-Reference Metrics

Mean Squared Error (MSE) Root Mean Sqaured Error (RMSE) Peak Signal-to-Noise Ratio (PSNR) Structural Similarity Index (SSIM) Universal Quality Image Index (UQI) Multi-scale Structural Similarity Index (MS-SSIM) Erreur Relative Globale Adimensionnelle de Synthèse (ERGAS) Spatial Correlation Coefficient (SCC) Relative Average Spectral Error (RASE) Spectral Angle Mapper (SAM) Spectral Distortion Index (D_lambda) Spatial Distortion Index (D_S) Quality with No Reference (QNR) Visual Information Fidelity (VIF) Block Sensitive - Peak Signal-to-Noise Ratio (PSNR-B)

No-Reference Metrics

Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) Natural Image Quality Evaluator (NIQE) Perception-based Image Quality Evaluator (PIQE)

For those metrics, we use the NRVQA repo.

Use ./image_metrics/calculate_metrics.py to calculate metrics. The program can handle a pair of images consisting of an uncompressed vs an compressed image.

$ python calculate_metrics.py birb.jpg  birb_compressed.jpg
using two images for comparison

Furthermore, two folders, consisting of either compressed and uncompressed images can be compared and their results averaged.

$ python calculate_metrics.py reference_images/ compressed_images/
using two directories with images for comparison

The output is a CSV file:

csv.out

With the following formatting:

image_name	rmse	psnr	rmse_sw	uqi	ssim	ergas	scc	rase	sam	msssim	vifp	psnrb	niqe	brisque
0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0