Implementation of a SqueezeNet[1]-type encoder and decoder for Semantic Segmentation in Tensorflow on the public Cityscapes dataset [2] for multi-GPU training.
Code structure is inspired by the Tensorflow Cifar10-tutorial: https://github.com/tensorflow/models/tree/master/tutorials/image/cifar10/
[1] Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, Kurt Keutzer; SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size; https://arxiv.org/pdf/1602.07360.pdf