This paper introduces our submission to the 2nd Facial Landmark Localisation Competition. We present a deep architecture to directly detect facial landmarks without using face detection as an initialization. The architecture consists of two stages, a Basic Landmark Prediction Stage and a Whole Landmark Regression Stage. At the former stage, given an input image, the basic landmarks of all faces are detected by a sub-network of landmark heatmap and affinity field prediction. At the latter stage, the coarse canonical face and the pose can be generated by a Pose Splitting Layer based on the visible basic landmarks. According to its pose, each canonical state is distributed to the corresponding branch of the shape regression sub-networks for the whole landmark detection. Experimental results show that our method obtains promising results on the 300-W dataset, and achieves superior performances over the baselines of the semi-frontal and the profile categories in this competition.