HMLC

Learnable Coding

Introduction

This repository contains our public tools for ....

Prerequisites

In order to compile and run the tools provided in this repository you will need:

1. Python 2.7
2. ffmpeg (version 2.8.15 or higher)

To produce data for training the network

To produce the necessary data to train the network run

python Produce_HM_InfoPerFrame.py --f=../vid/park_joy_640480.mp4

Option	Description [default]
--f	input file name
--qp	quantization parameter [30]
--mcu	maximum CU size [64]
--mpd	maximum partition depth [4]
--nf	number of frames to be encoded [900000000]
--fps	frames per second [30]
--path	path for results [../HMLC_Results]
--w	video width [640]
--h	video hight [480]
--rate	target bitrate [100k]
--cfg	HM configuration file [./encoder_lowdelay_P_main.cfg]

A successful run will produce the following:

• a binary file with number of bits required to encode each picture
• a folder contains the encoded pictures; Px.jpg is the fifth encoded picture
• a folder contains all inter-prediction pictures; Px_y.jpg is the reference picture y used for inter-prediction of picture x