README.md

metadata

license: cc-by-3.0

Architecture: Resnet-18 with two modifications.

1. 1 channel Conv2D as the first layer.
2. 2-way output on FC layer.

Training procedure:

1. Pre-trained in ImageNet.
2. Further training on FashionMNIST.
3. Final training on the task of predicting if Fashion-MNIST images are flipped vertically or not.