Inspiration:

Our school recently added a segment to assemblies in which individuals go up on stage and read a poem out loud in front of the school. During one assembly, two teachers improvised poems based on images chosen from the internet. This segment inspired us to create PhotoPoet, which uses machine learning to generate poetry based on the features of the user’s inputted image.

What it Does:

It’s a simple process: users upload a picture from their device to our website, and within a few seconds a poem pertaining to that picture is displayed on the screen below it.

How We Built It:

To build this project, we used Python to process the data and Keras (with backend Tensorflow) in Python to create and train the model. As for the front-end, HTML, CSS, and JavaScript were used to design and develop the functionality of our interface on the website. We also used node.js to host the site and use the image recognition API, as well as Flask to send the keywords generated by the API to our trained machine learning model. The model was trained using the GPU on Google Colab.

Accomplishments We Are Proud of:

We all agree that we worked very well together; we discussed matters both pertaining and not pertaining to the project with humour and amiability. As for the code in specific, finding an online cloud service to host the dataset so we could train it was difficult. Even more so was the time it took to train the model - some training sessions were projected to last multiple days. Although training our model for a few more hours (or days) would have yielded significant improvements to our project, we are proud to have create something that we thought was beyond our skill levels just a few weeks ago.

What We Learned:

We believe that throughout the 36 hours, we learned how to balance the work and stress of designing a software project with the fun activities and company around us. We also learned a ton about machine learning, knowledge that is super valuable and pertinent to the world we live in and the industries we would like to pursue in the future.

Built With

Share this project:
×

Updates