- Fashion MNIST dataset : https://www.kaggle.com/zalando-research/fashionmnist/data
- Benchmark dashboard : http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/#
Tried and tested :
- Stacked, deep CNN architectures (~90.5% test acc.)
- Stacked, deep CNN architectures with dropout, data augmentation or both (~92% test acc.)
To-Do :
- Usage of SVM loss instead of softmax loss
- Ensembles of deep CNN models with SVC/kNNs etc.
- Try application of pre-trained ImageNet models
- CapsNet implementation on Fashion MNIST