https://github.com/kyzn/minibatchimagemorphing

An attempt at introducing morphed in between images to mini batches while training a Deep Neural Network.
https://github.com/kyzn/minibatchimagemorphing

deep-learning machine-learning mnist neural-network ucsc

Last synced: 3 months ago
JSON representation

An attempt at introducing morphed in between images to mini batches while training a Deep Neural Network.

• Host: GitHub
• URL: https://github.com/kyzn/minibatchimagemorphing
• Owner: kyzn
• Created: 2016-06-11T18:38:07.000Z (about 10 years ago)
• Default Branch: main
• Last Pushed: 2020-06-13T19:05:57.000Z (about 6 years ago)
• Last Synced: 2025-12-26T11:43:45.217Z (7 months ago)
• Topics: deep-learning, machine-learning, mnist, neural-network, ucsc
• Language: Jupyter Notebook
• Size: 172 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Image Morphing for Mini Batches in DNN

![An example for morphing: each row is a different digit.](https://github.com/kyzn/MiniBatchImageMorphing/raw/master/0%20Example/Example.png)

As part of UCSC - CS218 "Deep Learning" class, we wanted to work on possible ways to improve our test accuracy. One idea was to introduce more images to mini-batches. We did it by creating new images. 

This code is based on [Michael Nielsen's code](https://github.com/mnielsen/neural-networks-and-deep-learning/blob/master/README.md), which uses MNIST dataset. We also used the same data.

To get the data file, please right click and save [mnist.pkl.gz](https://github.com/mnielsen/neural-networks-and-deep-learning/raw/master/data/mnist.pkl.gz).

## Results

|                     | Min Test Cost | Max Test Accuracy |

| ------------------- | ------------- | ----------------- |

| Benchmark           | **0.6108**    | **97.00%**        |

| With Image Morphing |   1.3728      |   95.07%          |

## License

Please visit https://github.com/mnielsen/neural-networks-and-deep-learning/blob/master/README.md for license details of *Code samples for "Neural Networks and Deep Learning"*.