tags: Deep Learning, Computer Vision, Segmentation, CVPR 2015
Reinterprete classification nets as fully convolutional and fine-tuning from their learned representations.