Welcome! We will now guide you in the setup and first run of Segue. Segue is a video preparation system that adapts video chunking depending on the expected rate adaptation and network behavior.

If you haven't yet, please read our paper published at JSys!

Install docker according to your OS and make sure your current user is able to use docker. In a default installation on Ubuntu Linux, only root is allowed to run docker containers. In this case, make sure to follow the instructions as root user.

Download the docker image (about 2 GB) with the following command:

docker pull melissalicc/maya

Clone or unpack Segue, open a terminal and change directory to the docker subdirectory: cd segue/docker. Then, enter the docker container with the run.sh command. This script uses some relative paths, so it is important to run it while being in the docker directory. (I.e don't do a ./docker/run.sh this will not work).

This script will start bash within the docker container, the segue directory will be mounted in /segue. So after you do a cd /segue you can invoke the execution script.

In ./configs all the details for configure Segue segmentation and augmentation strategies are explained. Also, the preset of configurations tested in our paper are made available.

To execute Segue, simply run the following command.

./execute_segue.sh <Vid> <Abr> <Tr> <Vmaf> <Gr> <Aug> {<R_Gr> <R_Aug}

The parameters description is the following:

• Vid: Refers to the video ID. Please map your video ID to its configuration file in make_run/LOC.py
• Abr: Refers to the ABR under test. The implemented ABRs are a rate-based (RB), buffer-based (BB), and two versions of robust MPC (RMPC-A for the version that uses VMAF, RMPC-O for the version that uses bitrate).
• Tr: Refers to the trace set under test. Map the trace set ID to their location in the file make_run/LOC.py.
• Vmaf: Refers to the VMAF model at which the evaluation should be run. Acceptable values are MOBILE/HDTV/4K.
• Gr: Refers to the grouping strategy utilised by Segue. A description of the accepted parameters is below.
  • K-{L}: Constant segmentation strategy. L can be either 1, 2, 3, 4 or 5.
  • GOP-{L}: GOP based segmentation strategy from related works. L can be either 1, 2, 3, 4 or 5.
  • TIME: Time based heuristic, referred in the paper as Time.
  • BYTES: Time based heuristic, referred in the paper as Bytes.
  • TIME+BYTES: Time based heuristic, referred in the paper as Time+Bytes.
  • SIM: Short term simulation based segmentation strategy, referred in the paper as Sim.
  • WIDE-EYE: Short term simulation based segmentation strategy, referred in the paper as WideEye.
• Aug: Refers to the augmentation strategy utilised by Segue. A description of the accepted parameters is below.
  • NONE: No augmentation is applied. This is the option to be utilised in case one wants to isolate and test the performance of segmentation only.
  • LAMBDA-B: Augmentation heuristic referred in the paper as $\lambda_b$.
  • LAMBDA-V: Augmentation heuristic referred in the paper as $\lambda_v$.
  • LAMBDA-BV-{B}-{V}: Augmentation heuristic referred in the paper as $\lambda_{bv}$. B and V refer to the bitrate and VMAF thresholds respectively. B value can be 5, 10 and 15, while V values are in the range of [5, 15[.
  • SIGMA-BV: Simulation based augmentation strategy referred in the paper as $\sigma_{bv}$.
  • CBF-{T}: CBF chunk selection strategy from related works. Threshold values T can be 40, 60 and 80.
  • SIVQ-{T}: SIVQ chunk selection strategy from related works. Threshold values T can be 5, 10 and 15.
• R_Gr, R_Aug (OPTIONAL): refer to the reference segmentation and augmentation strategy from which Segue performance has to be evaluate against. They follow the same parameter alphabet of Gr and Aug and, if not specified otherwise, they are set to K-5 and NONE.

If you would like, for example, to execute WideEye segmentation, absent of augmentation for video B and a buffer based algorithm for a trace set denominted SLOW with the VMAF MOBILE perceptual quality model, you would run:

./execute_segue.sh B BB SLOW MOBILE WIDE-EYE NONE

An example of output will be: