Inspiration
With my class in Machine Learning, CSE 4309, my amazing professor has given the class a good understanding of neural networks. She has given the class an understanding of how to train a CNN for image data.
What it does
It trains on the training data, validates the model with the validation set, then gives out the results of the test. It does this through training of a CNN.
How we built it
Using Keras along with many other libraries
Challenges we ran into
There are over 100k images in the train set, and each image is 200x200 as well as being in colour, so it becomes 200x200x3 floats for a single picture. This means at (about) 150k pictures just for training, there would be 18,000,000,000. So I would need 18 billion floats. So in order to make this work, when training the CNN with batches, the images were loaded in and normalized as they were needed.
Accomplishments that we're proud of
That it is at least predicting the results, somewhat
What we learned
Dealing with model training for such large amounts of data needs the data loaded at the time of prediction
What's next for the project
Potentially seeing if I can get a better accuracy
Built With
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.