Fall 2023

DropShop

DROPSHOP

CONTENTS

Fall 2023

DropShop

DIVING INTO RESEARCH & IMPLEMENTATION

INTRODUCTION

INITIAL CHALLENGES

QUARTER GOALS

ROADBLOCKS

01

02

03

04

LESSONS LEARNED

PERSONAS

DEMO / FINAL REPORTS

05

06

07

INTRODUCTION

• Professor Hui’s lab uses a program called “Dropshop” to deal with the interactions between droplets in Microfluidic chips.
• Computer vision system focus
• Code created by prior PhD student Erik Werner
• Issues with droplet disappearance, no case for handling disappearing droplets

​

​

Prior Code

Project context

A

Fall 2023

Introduction

• As a team we decided to focus on improving the COMPUTER VISION SYSTEM by adding the ability for the droplet to be relabeled.

​

Our Team

Accuracy, Reliability, & Usability

INITIAL CHALLENGES

• System cannot keep with up with fast generation of Droplets

Accurate labels for the system

​

​

Fall 2023

Challenges

Current CV tracking could not keep up with droplet generation, some droplets are renamed as it traverses the channel when it should be the same droplet.

• Droplets disappearing would be reacquired with a new-id

Real-time processing

​

​

• Ensuring the system operates in real-time while maintaining high accuracy

Droplet Interaction

​

​

• Developing an object tracking system capable of handling complex scenarios such as droplet collisions, merges, and splits.

User-interface

​

​

• Developing a user friendly interface that allows users (even those unfamiliar with code) of DropShop to easily interact with

QUARTER GOALS

• Label the droplets with an ID in order of the droplets being dispense

Labeled object detection model

​

​

Fall 2023

Goals

Handle droplet detection disappearances

​

​

• Simple interface for the user to create bounding boxes to track the droplets.

​

User interface

​

​

• When droplets disappear keep the same ID

Data analysis/comparisons to previous versions

​

​

​

​

• Analysis of our code compared to Erik Werner's based on decided metrics.

ROAD BLOCKS

• In order to improve the machine learning model need to improve data set with data labeling

Limited set of data

​

​

Fall 2023

Road Blocks

Prior Code

​

​

• Delegating tasks between Team Members was for the initial weeks was difficult and we struggled to establish work parallelism.

​

Adapting to an Agile work process

​

​

​

​

• Code was extensively large and written in C++ which we did not have extensive working knowledge with

Initial Project Scope

• Since the prior code was based on a PHD student dissertation, determining what we could’ve improved for the project was difficult.

​

LESSONS LEARNED

• Importance of clear communication channels and in-person meetings
• Clear division of tasks with SCRUM task
• Using SCRUM techniques such as product backlog helps gauge completion and progress.
• The use of JIRA for sprints is a tool to assist SCRUM
• Using Work parallelism allows the ability to finish multiple tasks between team members

Agile process

​

​

​

​

Fall 2023

Lessons Learned

ML Model and Data Labeling

​

​

​

​

• Learning use of YoloV, Ultralytics, and Roboflow
• Delegating the task of Data Labeling between multiple Team Members
• The process of accurately annotating microfluidic droplets for object detection can be time-consuming

PERSONAS

• Optimized and error-free software to track droplets for biological testing/ research

Researchers

(MOST TO LEAST ESSENTIAL)

Fall 2023

Client

• Include code comments, explanations, and diagrams to make the documentation comprehensive and easy to understand for non-technical users.
• Store the documentation in a version-controlled repository for easy access and updates as the project evolves.

​

Engineers

• Implement efficient data preprocessing and post-processing steps to improve the overall speed and accuracy of the system.
• Regularly monitor and evaluate the model's performance to ensure it meets the accuracy requirements for your business promises.

​

Business Partners

Accuracy, Reliability, & Usability

Fall 2023

Final Reports

Final droplet order: [34, 25, 29, 35, 39, 41, 44, 47]

Droplets were lost and relabeled very frequently

Droplet Tracking

Processing Time

DROPSHOP LAI’S VERSION

Total Runtime: ~13 minutes

Fall 2023

Final Reports

DROPSHOP CURRENT VERSION

Processing Time

Total Runtime: ~50 seconds

Fall 2023

Final Reports

Droplet Tracking

Final droplet order: [7, 1, 2, 3, 8, 4, 5, 6]

DROPSHOP CURRENT VERSION

Fall 2023

Demo

INTERFACE DEMO

TRAINING DEMO

Fall 2023

Demo

Fall 2023

Demo

TEST DEMO

Fall 2023

Special Thanks

THE LAB STAFF, PROF. HUI, & PROF. NAWAB

​

A SPECIAL

FOR THEIR INVALUABLE GUIDANCE DURING THIS PROCESS

GENERAL Q&A

• Value of clear communication channels and in-person meetings
• Value of meeting with an objective third party to explain our work at a high level was good exposure to speaking with real-world clients
• Having flexibility in a client is equally a blessing and curse when trying to define initial goals for the project

​

Top 3 Takeaways

​

​

​

​

Fall 2023

General Q&A

Teamwork Experience

​

​

​

​

• Consolidating communication to a single channel/ method for the team early on helped reduce miscommunication. Relied heavily on Slack for class and project partner logistics and text messaging for team communication
• Managing different personalities and work ethics is easier when you delegate time to connect in person (even if it’s just at the beginning stages)

Agile/ SCRUM

• Continuous delivery is a way to deliver a product and receive consistent feedback to make quick changes as a client desires
• Setting a hard deadline of when to move on from failed ideas helps move product development forward