img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation
This work provides a more efficient and informative approach to face alignment and detection for computer vision researchers and developers by directly estimating 6DoF pose.
This paper proposes a real-time method for 6DoF 3D face pose estimation that does not require prior face detection or landmark localization. The method outperforms state-of-the-art face pose estimators on AFLW2000-3D and BIWI datasets, and surprisingly, also surpasses comparable SOTA models on the WIDER FACE detection benchmark, despite not being optimized for bounding box labels.
We propose real-time, six degrees of freedom (6DoF), 3D face pose estimation without face detection or landmark localization. We observe that estimating the 6DoF rigid transformation of a face is a simpler problem than facial landmark detection, often used for 3D face alignment. In addition, 6DoF offers more information than face bounding box labels. We leverage these observations to make multiple contributions: (a) We describe an easily trained, efficient, Faster R-CNN--based model which regresses 6DoF pose for all faces in the photo, without preliminary face detection. (b) We explain how pose is converted and kept consistent between the input photo and arbitrary crops created while training and evaluating our model. (c) Finally, we show how face poses can replace detection bounding box training labels. Tests on AFLW2000-3D and BIWI show that our method runs at real-time and outperforms state of the art (SotA) face pose estimators. Remarkably, our method also surpasses SotA models of comparable complexity on the WIDER FACE detection benchmark, despite not been optimized on bounding box labels.