-
-
A look at our training
-
Our Chatbot (though limited as of right now!)
-
The output
-
The base page
Inspiration
Every member on our team is a massive fan of dogs. A couple of us have had dogs that passed away and know what it feels like to lose them. We wanted to create a product that would be able to help monitor their health and make their lives easier in any way possible. VetView allows dog owners to keep a check on their dog's health by inspecting their stools without needing to manually handle it.
What it does
VetView asks the user to input the breed and age of their dog alongside the picture of their stool, and gives information based on these details, such as what their stool quality might indicate about their gut health, why any deterioration might have occurred, and also has a Chatbot that provides in-depth advice on how to improve their gut health.
How we built it
The image that is input by the user is fed into an image classification model that was trained by us. This model was trained using an image dataset that we have found on github for dog stools. This dataset had 4 classes: normal, soft, lack of water, and diarrhoea. We used transfer learning with EfficientNetB0 as our base and added additional layers on top of it. The class predicted by the model is used to give a rudimentary analysis on the dog's condition using research that we did ourselves. This class is also used in conjunction with the age and breed of the dog (input by the user) to prompt ChatGPT via the OpenAI API into giving a more in-depth analysis. Our front-end and back-end is built using Flask, HTML, Javascript and CSS.
Challenges we ran into
The first challenge, and probably the biggest one was finding an appropriate dataset that had enough images and that was of high quality. It took us a fair amount of time to come across the dataset we used, and even longer to sift through it and reclassify any misclassified images, or remove any low quality images. The second challenge was integrating the two halves of our project (the Image Classification and Chatbot) while maintaining our front-end and back-end. There were several times when our frameworks would clash leading to errors we have never encountered before. Getting through this was a massive team effort.
Accomplishments that we're proud of
We are proud of making a functional, working product that does not give any errors, classifies and gives appropriate information over 90% of the time, and one that has a very real chance to have an impact on the pet health industry. The framework we have come up with is easily scalable, and would be even more effectively implemented on an app that allows users to directly take pictures of the stools.
What we learned
We learned several new techniques and technologies. Every team member had to work on a software aspect that they had never used before. Here's some individual feedback! Kalyan: I learned how to make a Chatbot and integrate a model into the backend of a website Kaushik: I learned how to master my version control skills and integrate the front-end with the back-end Kabir: I learned how to build a transfer learning model and how to best sanitize a dataset for training the model. Varnica: I learned how to use version control skills and how to debug code efficiently.
What's next for Vet View
VetView has incredible potential to grow and scale across platforms. 1) VetView as an app would be optimal. Currently our team does not have enough experience in app development, so we thought it wise to create a website for VetView to begin with. However, allowing the user to directly click and upload a picture of their dog's stools is extremely convenient. Additionally, each user can have an account that would help track their dog's health as the app can also save the pictures uploaded by the user across a period of time. 2) Due to a lack of time and resources, our dataset only had 4 classes, when most vets generally classify dog stools into 7 classes. Given more time we could either find a better dataset with 7 classes, or data scrape ourselves which would vastly improve our performance too. We can also make our Chatbot more interactive given more time. 3) Finally, VetView can be scaled by extending the product to other pets such as cats and birds.
Log in or sign up for Devpost to join the conversation.