Reviews: Learning Deep Structured Multi-Scale Features using Attention-Gated CRFs for Contour Prediction

This paper proposes a gating mechanism to combine features from different levels in a CNN for the task of contour detection. The paper builds upon recent advances in using graphical models with CNN architectures [5,39] and augments these with attention. The paper also presents an ablation study where they analyze the impact of different parts of their architecture. Cons: 1. Unclear relationship to past works which use CRFs with CNNs [5,39] and other works such as [A,B] which express CRF inference as CNNs. The paper says it is inspired from [5,39] but does not describe the points of difference from [5, 39].