GivingWorldItsLostVision

• 

  HuggingFace App

• 

  Creation Pipeline

• 

  Plots

• 

  Github

• 

  Local Install readme

Inspiration

My grandma got operated for cataract this year, while cataract is an age related condition other eye related retinal diseases might be predicted by the use of computer vision and further be prevented or cured. So, AI helps in making early detection and identification a possibility.

What it does

The application takes in an image of an OCT scan and identifies the related retinal disease if it exists from the 3 categories of CNV, DME or drusen or it is a normal condition

How we built it

It was built by identifying a suitable dataset from Kaggle
Then, the architecture to be implemented was identified
Once, architecture was identified and implemented the model was trained
After training, proper testing was done
Plots were made to see the training
App interface was made and model deployed

Challenges we ran into

The model was overfitting with low accuracy so, used a little augmentation and scheduling to improve those
Model underfitting too much
Limited GPU Access for model training so cant train for too long
Gradio being down for a good time postponing deployment to the last second

Accomplishments that we're proud of

A validation accuracy of 94% and testing accuracy of 88%
Fast model prediction

What we learned

Handling and tackling underfitting and overfitting
Making an app interface to a machine learning problem

What's next for GivingWorldItsLostVision

Testing out on more models and datasets to improve it even further

Built With

• canva
• github
• gradio
• huggingface
• python
• pytorch