Various deep learning models in torch
- DeepLab model (DeepLab-MSc-COCO-LargeFOV) without CRF for post-processing. Consists of only the model structure for training. Based on caffe implementation at http://ccvl.stat.ucla.edu/ccvl/DeepLab-MSc-COCO-LargeFOV/train.prototxt Features exploited: (1) Multiscale features (2) extra annotations from MS-COCO dataset (pretrained on MS-COCO, and then fine-tuned on PASCAL) (3) large Field-Of-View.