[IEEE TCSVT'25] MERINA+: Improving Generalization for Neural Video Adaptation via Information-Theoretic Meta-Reinforcement Learning

by Nuowen Kan, Chenglin Li, Yuankun Jiang, Wenrui Dai, Junni Zou, Hongkai Xiong, and Laura Toni at Shanghai Jiao Tong University

This repository is the official Pytorch implementation of MERINA_Plus

Abstract

This document illustrates how to obtain the results shown in the paper.

Code Structure

The repository is organized as follows:

• main.py: The main entry point for running the experiments and training the models.
• algos/: Contains the core algorithm implementations, including reinforcement learning agents (e.g., PPO), variational autoencoders (VAE), and training scripts for video adaptation.
• baselines/: Implements baseline algorithms for comparison, such as MPC, Bola, and other ABR methods.
• config/: Configuration files, including argument parsers, data sources, and labels for experiments.
• envs/: Environment definitions for video streaming simulations, including video size and VMAF data.
• utils/: Utility functions for plotting, data loading, and result visualization.
• requirements.txt: Lists the Python dependencies required for the project.
• torch.yaml: Conda environment file for easy setup of the PyTorch environment.

The environment setup

Anaconda is suggested to be installed to manage the test environments.

Prerequisites

• Linux or macOS
• Python >=3.6
• Pytorch >= 1.6.0
• numpy, pandas
• tqdm
• seaborn
• matplotlib
• CPU or NVIDIA GPU + CUDA CuDNN

Install PyTorch. Note that the command of PyTorch intallation depends on the actual compute platform of your own computer, and you can choose appropriate version following the guide page. For example, if you have intalled CUDA 10.2, you can intall PyTorch with the latest version by running this Command:

conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

Or you can create a specific environment (many redundant dependencies will be installed) just via

conda env create -f torch.yaml