Convolutional Neural Networks (Computer vision)
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.ipynb_checkpoints
MNIST (sota)
assets
dogs_cats
LICENSE
README.md
resnet.py

README.md

Convnets


State of the art MNIST

A CNN model which achieves 99.7% accuracy. It makes use of Vgg16 model. I've used data augmentation, batch normalization, dropout, maxpooling . The model has been finetuned. Accuracy on training set = 99.42 and accuracy on validation set = 99.45. The model doesn't overfit at all. Three models have been used:

  • Linear model
  • Single dense layers
  • Vgg style CNN

Requirements

  • keras==1.2.2
  • Python 2.7
Use model.summary() to gather more intuiton about convolutional layers

mnist_epoch

Dogs and cats classification

Made use of Vgg16 model(which won Imagenet competition in 2014).

Models has been finetuned in order to classify image as dog or cat.

Download data:

dog


Deep Residual Network

ConvNet architecture with Skip connections. Can make ConvNet more deeper with increased acccuracy.

Acquiring this repo

$ cd ~
$ git clone https://github.com/Convnets
$ cd machine-learning-programs

Contributions

Contributions are always welcome in the form of pull requests with explanatory comments. You can Fork this repo and send in pull request to expand the repository if they can fit into this repository.