DeepLab-Attention-COCO employs (1) multi-scale inputs, (2) extra supervision, and (3) attention model. The model has been pretrained on MS-COCO dataset.
After DenseCRF, the model yields 75.7% performance on PASCAL VOC 2012 test set.
CRF parameters: bi_w = 4, bi_xy_std = 20, bi_rgb_std = 3, pos_w = 3, pos_xy_std = 3.
Pretrained models and corresponding prototxt files
Please download from this link.