This pandemic taught us the importance of managing social gatherings. We wanted to develop this application that can aid authorities in managing crowds by using crowd sourced crowd images.
What it does
Enumerate is a crowd counting platform on which users can upload images of crowds and the platform in return will display the estimated count of people in the image and also display the heatmap of crowd density.
How we built it
We had to develop a Machine Learning model capable of counting the size of the crowd in an image. For this, we implemented a dilated CNN which generates a crowd density heatmap for a given crowd image. This model was proposed by Li et. al in their paper CSRNet.
We trained this model on ShagaiTech dataset (CVPR 2016) and tried to optimize our model using MSE Loss.
Challenges we ran into
Crowd counting was a new topic for us. We had to read up literature on this topic and implement models in under 24 hours. We found this exhilarating. Although it was very time consuming to train the models
Accomplishments that we're proud of
We successfully trained a ML model and deployed it
What we learned
Crowd counting was a problem statement that was totally new for us. We thoroughly enjoyed working on this problem statement and
What's next for Enumerate
Given the current pandemic-scenario, we believe a crowd counting platform like Enumerate will be a really useful tool to control and manage crowds. We plan to open source this code and the trained model for people to use.