Gaussian Conditional Random Field Network for Semantic Segmentation - Supplementary Material

Notations: We use bold face small letters to denote vectors and bold face capital letters to denote matrices. We use A>, A−1, |A| and trace(A) to denote the transpose, inverse, determinant and trace of a matrix A, respectively. We use ‖b‖2 to denote the squared `2 norm of a vector b. A 0 means A is symmetric and positive semidefinite. We use R to denote the set of real numbers and E to denote expectation. 1. Mean field inference In this work, we model the conditional probability density P (y|X) as a Gaussian distribution given by