nordic96/infinite_challenge

Challenge is Infinite

Hi there, I'm Stephen 👋

I'm Stephen, a computer science graduate from NUS, Singapore. I am currently working as a Frontend Developer, and I am very excited to pursue my career in Frontend Engineering!

• 🔭 I’m currently working on Frontend Development
• 🌱 I’m currently learning ReactJS, Typescript, Redux, and many more!
• 👯 I’m looking to collaborate on Establishing a structured testing environment for React based UI applications

Stats Overview

Language Used Overview

infinite_challenge's Issues

Facial Recognition for all members with good accuracy

Modularize pipeline.py

[DOCKER] Useful Gists

1. Install Docker Desktop (Windows/Mac)
2. Try build docker image using our Dockerfile LOCALLY
   • Open CLI (iTerms/WSL)
   • cd your_path_to_project_dir/Infinite_Challenge
   • docker build --tag infinite_challenge:0.0.1 . Building the docker image
   • docker run --publish 8000:8080 --detach --name any_alias_youwant infinite_challenge:0.0.1 Running your docker image locally
3. Pulling from Docker Hub docker pull nordic96/infinite_challenge:latest
4. Pushing to new tag to docker hub docker push nordic96/infinite_challenge:tag_name
5. Removing Docker image in LOCAL docker rmi [image id]
   • Docker image ids can be listed using this command docker images
   • Docker containers can be listed using the following command docker ps -l

[Reference]
Build and run your image: https://docs.docker.com/get-started/part2/

Skull detection in frame

• Skull Detection

• Information Update in CSV

Improve skull detection accuracy

Current yolo model produces false positives in multiple test videos

Improve facial recognition (adding 1 member)

1 missing member for facial recognition ('Gil')

Log File implementation for Jobs

Modify strings.config to .ini file

.ini is the correct format for python files

S3 upload implementation

Connecting the pipelines

Phase1 and Phase 2 needs to be connected
in Phase 1:

• video processing using skull detection
• save screenshot if skull is detected & create a csv with skull info (skull location(s), timestamp)

in Phase 2:

• facial recognition from saved screenshots
• update the created csv file from Phase 1 (member who is detected)

Finalised csv file should be ready #12 for db creation

Remote deployment

Things we probably need to look at

• Ability to run code on a remote virtual machine(s)
• Cloud storage for input video files and extracted images (capacity may be an issue)
• Job scheduling and to prevent overlapping (processing same input file) if parallelization is implemented
• Server-client model for updating csv if multiple processes are accessing the same file to prevent errors

Skull & Face shortest distance estimation

Better automation of parralel jobs

• Job management for if parallel jobs are run (not as a AWS Batch Array Job)
• Separating each phase of the pipeline so resources can be better distributed
• Running a partially complete job (which failed midway during the pipeline)

Remote Deployment

Things we probably need to look at

• Ability to run code on a remote virtual machine(s)
• Cloud storage for input video files and extracted images
  X Job scheduling and to prevent overlapping (processing same input file) if parallelization is implemented
  X Server-client model for updating csv if multiple processes are accessing the same file to prevent errors

Log Frame info. in CSV

Video processing for each frame

util function
Log information of the frame in CSV for easier hand-over to Facial recognition model.
Logging includes:

• Timestamp of the frame in video
• Members detected in the frame
• Episode no. (filename of the directory)

Config file module

configurations and default parameters will be managed in strings.config

Inserting CSV data into SQL server

Suggesting the following implementation:

• Bulk insert into RDS after job is complete
• Check episode information before processing episode

Project Documentation

• readme

Creating Docker Container for AWS Batch