DeepLab is the basic model we employed in our ICLR 2015 paper (Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs). We employed kernel size = 4×4 at the first fully connected layer of VGG-16 and input stride = 4, resulting in a receptive filed size of 128.
Performance
After DenseCRF, the model yields 66.4% performance on the PASCAL VOC 2012 test set.
CRF parameters: bi_w = 5, bi_xy_std = 50, bi_rgb_std = 10, pos_w = 3, pos_xy_std = 3.
Pretrained models and corresponding prototxt files
Please download from here