S2F2: Self-Supervised High Fidelity Face Reconstruction from Monocular Image
This work addresses the need for high-fidelity face reconstruction in computer vision applications, though it appears incremental by building on existing auto-encoder and ray-tracing methods.
The paper tackles the problem of reconstructing detailed face geometry and reflectance from a single monocular image, achieving high-fidelity results with self-supervised learning and real-time computational speed.
We present a novel face reconstruction method capable of reconstructing detailed face geometry, spatially varying face reflectance from a single monocular image. We build our work upon the recent advances of DNN-based auto-encoders with differentiable ray tracing image formation, trained in self-supervised manner. While providing the advantage of learning-based approaches and real-time reconstruction, the latter methods lacked fidelity. In this work, we achieve, for the first time, high fidelity face reconstruction using self-supervised learning only. Our novel coarse-to-fine deep architecture allows us to solve the challenging problem of decoupling face reflectance from geometry using a single image, at high computational speed. Compared to state-of-the-art methods, our method achieves more visually appealing reconstruction.