Facial Landmark, Head Pose, and Occlusion Analysis using Multitask Stacked Hourglass

In this study, we proposed a multitask network architecture for three attributes, landmark, head pose, and occlusion, from face image. A 2-stacked hourglass with task-specific heads is the architecture. We also designed auxiliary components network. First feature pyramid fusion module, which plays crucial role in facilitating contextual information various receptive fields. Second interlevel occlusion-aware explicitly fuses intermediate occlusion prediction between subnetworks. The third gimbal-lock-free pose head, outputs rotation matrix 6D representation. conducted an ablative study of these to determine their impacts on Additionally, introduced landmark heatmap scaling approach avoid falling local minima. trained 300W-LP dataset C-CM occlusion. Then, fine-tuned using 300W or WFLW dataset, instead task. This 2-stage training method contributes enhancing detection accuracy that other tasks. experiments, assessed eight test datasets metrics.