codec_compare

A small framework for comparing metrics for still image codecs. The framework is portable, and runs inside a Docker container. The Dockerfile includes 4 anchor codecs (jpeg, kakadu, webp, hevc) and encoding/decoding scripts for each codec. By default, the framework encodes 8 BPP targets per encoder (0.06, 0.12, 0.25, 0.50, 0.75, 1.00, 1.50, 2.00) and two colorspaces (RGB and YUV420). The encode and decode for each codec is handled by small scripts placed in the ./encode and ./decode directories. The anchor codec scripts are written in Python, but you could use any language you'd like. Each script is called by the framework with the following command-line arguments:

Encode args:

image_src  = sys.argv[1]
image_out  = sys.argv[2]
bpp_target = sys.argv[3]
width      = sys.argv[4]
height     = sys.argv[5]
pix_fmt    = sys.argv[6] 
depth      = sys.argv[7] 

Decode args:

image_src  = sys.argv[1]
image_out  = sys.argv[2]
width      = sys.argv[3]
height     = sys.argv[4]
pix_fmt    = sys.argv[5] 
depth      = sys.argv[6]

Source images:

Place your source images in ./images_classes/class<X>_<bitdepth>bit/ for classes A and B, Example: ./images_classes/classA_8bit/. Note: .yuv files in class A 10 bit cannot be handled in this version!

Place your source images in ./images_classes/class<X>/ for classes C, D and E. Example: ./images_classes/classC/.

Place your source images in ./images_classes/classE_exr/ for class E images in .exr format

To add another codec:

Update the Dockerfile to include your binaries. Add an encode and decode script in ./encode and ./decode.

To build container:

docker build -t codec_compare .

To run container:

docker run -it -v $(pwd):/codec_compare codec_compare

To encode, decode, calculate metrics:

./compare.py <path to images folder>/ Example: ./compare.py images_classes/classA_8bit/

Notes from PINAR:

If you want to exclude a codec, remove the .py file from both ./encode and ./decode folders.

If you are going to encode and decode images with bit depth greater than 10bpp, after building the container you have to enable RExt__HIGH_BIT_DEPTH_SUPPORT. Here's how:

After you build and run the container, type the following commands:

apt-get update

apt-get install vim

Y

vi /tools/HM-16.18+SCM-8.7/source/Lib/TLibCommon/TypeDef.h

In line 132 you have #define RExt__HIGH_BIT_DEPTH_SUPPORT set to 0. You need to change this to 1, save and quit (press i to switch to insert mode, when you finish changes press ESC to exit insert mode, then type :wq and hit ENTER. Then build HM again.

Now you won't have problems.

And that's all :)

To generate graphs:

./visualize.py ./metrics/*.json