# Capstone Meeting - 02/08/2020

## Agenda
1. Matthew + Yick to discuss their updates from past month development involving:
* Dataset Processing
* Model Development
* Application Deployment
* [Note]: Would be good if we could add video links, or images to make it clearer

## Yick Updates
* raw data preprocessing pipeline to produce the training and labelling text as inputs to our model;
* integration and unit testing;
* in openpose, make sure yoy set the "params['process_real_time'] = False", otherwise, a lot of frames will be dropped to keep up with the fps from the input side, which is undesirable and consequential;

## Matthew Updates
* Developed a simple LSTM model in keras
* LSTM model acted as a basis for Yick to work on
* Rolling Window Data Structure
* Earliest keypoints as first index and latest points as last index
* Yick helped in pointing out a bug
* Recorded videos of myself as datasets
* Explain architecture through diagram
* 
* Deployed application over 'eventlet' and over HTTPS, which means that we can access camera on our own personal PC browsers (as long as you are connected on cisco) --> Good for demo

## What has been done (summary)
* Signs are focussed on emergency/help related signs
* Basic LSTM Model (unrefined)
* Automation in dataset processing (by Yick)
* Web application deployed onto VM

## What needs to be worked on?
* DATASET collection!! (more data needed)
* improve data preprocessing pipeline:
* incorporate data validation pipeline into the preprocessing;
* incorporate small gaussian noise on the video to:
* to generate more data;
* add variance
* Need to invest more into researching model
* What hyperparameters to use? HYPERPARAMETERS TUNING!
* What architecture?
* Can use datasets as .txt files (from Yick) and use it to experiment different models and architectures

## Assignment for leaders for the project: (To be Discussed):
* Application Development and Deployment: **Matthew**
* Dataset Processing and Unit Testing: **Yick**
* Model Development and Research: **Tsz Kiu**

## Steps Moving Forward
* Role assignments for new tasks
* How do we record video? What criterias?
* Research into model development
* Regularly update trello so we can track progress
* Meet up with Jonathan showing a demo of our app/model