Inspiration
We were inspired to create Candor Canvas because emotion detection is a valuable field to dive into, especially computer vision. We recognize that mental health, despite gaining acknowledgement in the past, still is widely ignored. For some people with speech impediments or who do not want to talk, body language can reveal hidden insights not seen before. We hope that Candor Canvas can revolutionize the mental health field by introducing computer vision in it.
What it does
Candor Canvas is a dynamic mental health tool for therapeutic institutions and organizations. When recording sessions, simply import an mp4 file into the website and us analyze the way your patient feels. We offer insights that you can use to discover opportunities to get to know your patient better.
How we built it
We utilized the fer2013 dataset, of about 34800 images spanning 7 emotions, to create an accurate model that predicts emotions on video. We used cv2 VideoCamera to display it on a website. Streamlit was the website deployer which we hosted the model on, added a browse files option to import any mp4 file to analyze.
Challenges we ran into
Our model did not have enough data, we used data augmentation to solve it. VS Code was not syncing to github, we made a new repository. Streamlit was not deploying. When it did deploy, it was only local to the user, could work on getting a website up. Unfortunately, the frame rate is low which is due to hardware limitations, we will work on improving that.
Accomplishments that we're proud of
We are proud of creating an accurate model with 67% accuracy. (Emotion detection is difficult in computer vision), and successfully getting a web page up. We recognize that this is a tool useful for mental health hospitals and therapists to understand their patients better.
What we learned
We learned to use Data Augmentation to make artificial data to improve the data. We learned about streamlit which is a great way to host models for free.
What's next for Candor Canvas
Candor Canvas is working on analyzing emotions from different people simultaneously to work with couples/marriage therapy.
We also plan initiate speech based analysis alongside our emotion detection algorithm to furthur improve relability.
While the model is accurate, We are working on ensembling it by training various models and combining them to furthur improve accuracy. We are also looking for uses on Criminal investigation by Candor Canvas. We believe emotions can reveal a lot about a person.
Log in or sign up for Devpost to join the conversation.