mnist-competition's Introduction

MNIST Competition Tensorflow KR Group

MNIST competition submission files
Used Keras
Model Architecture
- ResNet
- VggNet
- VggNet5

Performance

Model	Description	Accuracy
VGG-like	VGGNet-like but smaller	99.71%
Resnet-like	ResNet-like but smaller	99.60%
VGG-like	VGGNet-like but even smaller than the first one	99.63%
Final	Ensemble 3 models + Voting	99.80%

Run

Evaluation

python evaluation.py

Train

python resnet.py 10 # 10 epochs & resnet
python vgg16.py 10 # 10 epochs & vgg
python vgg5.py 10 # 10 epochs & vgg

File descriptions

├── evaluation.py # evaluation.py
├── images # model architectures
│   ├── resnet.png
│   ├── vggnet5.png
│   └── vggnet.png
├── MNIST # mnist data (not included in this repo)
│   ├── t10k-images-idx3-ubyte.gz
│   ├── t10k-labels-idx1-ubyte.gz
│   ├── train-images-idx3-ubyte.gz
│   └── train-labels-idx1-ubyte.gz
├── model # model weights
│   ├── resnet.h5
│   ├── vggnet5.h5
│   └── vggnet.h5
├── model.py # base model interface
├── README.md
├── utils.py # helper functions
├── resnet.py
├── vgg16.py
└── vgg5.py

Model Architectures

ResNet

VggNet

VggNet5

mnist-competition's People

Contributors

Stargazers

Watchers

mnist-competition's Issues

Corrupted h5 file

Hi, I think your model weights are corrupted.

I ran the h5debug command-line utility on the .h5 weight files in your folder model and I get an error that the files cannot be opened!

Indeed, the function load_models() in 'evaluation.py', which attempts to open the .h5 weight files in your folder model, fails to open them.

How to ensemble these three models?

In performance, there is a final result that nsemble 3 models and voting, could you provide detailed introduction?

No dropout (in VGG16)

I downloaded the 'vggnet.h5' from this_repo, I loaded and checked the summary it was not having any dropout layer. Why didn't you use it? any particular reason?

Recommend Projects