Skip to content

videoturingtest/VIPlatform

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 

Repository files navigation

VIPlatform

Welcome to the VI Platform. The VI Platform is run through a series of docker containers which operate in tandem to provide answers to questions related to popular drama's such as 'Friends'.

Container Setup

In order to set up the foundation of the VIPlatform, first get the ViPlatform.tar.gz file currently stored on the vic server. This contains eight tar files which are exports of the docker images needed to run the Platform.

When you have this file on your desired server, extract the tar.gz file and change into the correct directory.

tar xf ViPlatform.tar.gz
cd ViPlatform

In order to setup the docker images on your local machine, please run the following script with appropriate user privileges. WARNING! This script may take some time to run and requires 200Gb of free disk space.

sh vpp_commands.sh

To verify the script setup correctly run:

docker ps

and check that the 8 required containers have been installed and are running correctly

VI Platform Usage

The VI platform is run through 8 different docker containers. The 'container_login.sh' script should be contained in the 'ViPlatform/' directory you setup in the last step. This can be used to login to the required containers.

Main

The main docker container coordinates all tasks involved in responding to a VI question. This main process MUST be run before any of the other docker processes. In order to login to the main docker container, run:

./container_login.sh main

Next, in order to start this docker container's process; run:

cd /home/main
python3 demo_main.py

STT

The STT docker container is involved in processing questions. In order to login to this docker container, run:

./container_login.sh pio_test2

Next, in order to start this docker container's process; run:

cd /root/alta
python2 vtt_stt.py

Yonsei Extraction

The Yonsei Extraction docker container is involved in language processing. This container loads BERT (https://github.com/google-research/bert), so this container's process may take some time to start. In order to login to this docker container, run:

./container_login.sh yonsei_vtt

Next, in order to start this docker container's process; run:

cd /home/pytorch-template-master
python3 infer.py

Level-Classification

The Level-Classification docker container assigns a difficulty score to a given question for VI. In order to login to this docker container, run:

./container_login.sh level_classification

Next, in order to start this docker container's process; run:

cd /home/VTT/Level_Classification
/opt/conda/bin/python3 client.py  

KnowledgeBase

The KnowledgeBase docker container provides an answer to a given VI question. In order to login to this docker container, run:

./container_login.sh kbqa0.3

Next, in order to start this docker container's process; run:

cd /home
python ibricks.py  

SNU_Low

The SNU_Low docker container provides answers to VI questions which are classified with a low level of difficulty. In order to login to this docker container, run:

./container_login.sh snu_low

Next, in order to start this docker container's process; run:

cd /root
python3 client.py 

SNU_High

The SNU_High docker container provides answers to VI questions which are classified with a high level of difficulty. In order to login to this docker container, run:

./container_login.sh snu_vqa

Next, in order to start this docker container's process; run:

cd vtt_qa_pipeline/startup
python3 cli.py infer

AnswerSelection

The AnswerSelection docker container chooses between multiple answers to a given VI question. In order to login to this docker container, run:

./container_login.sh khu_answer_2

Next, in order to start this docker container's process; run:

cd workspace/Answer_Selection
/opt/conda/bin/python3 predict.py

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published