DeepLab-LargeFOV-Semi-EM-Fixed is trained on PASCAL using mixed annotations (some strong pixel-level labels and many weak image-level labels) and the Fixed-Bias EM algorithm (please see our paper for details). The model is based on DeepLab-LargeFOV.
After DenseCRF, the model (trained with 1.4K strong labels) yields 64.6% performance on the PASCAL VOC 2012 val set, and the model (trained with 2.9K strong labels) yields 68.5% performance on the PASCAL VOC 2012 test set.
CRF parameters: bi_w = 6, bi_x_std = 183, bi_r_std = 5, pos_w = 3, pos_x_std = 3
Pretrained models and corresponding prototxt files
Please download from this link.