Learning 3D-aware Image Synthesis with Unknown Pose Distribution
This addresses a bottleneck in 3D-aware image synthesis for computer vision and graphics researchers by enabling high-quality synthesis without pose priors, though it appears incremental as it builds on existing generative radiance fields.
The paper tackles the problem of 3D-aware image synthesis by eliminating the need for pre-estimated 3D pose priors, which can cause faulty geometry, and proposes PoF3D, a method that jointly trains a pose-free generator and pose-aware discriminator to automatically infer poses, achieving performance on par with state-of-the-art in image and geometry quality.
Existing methods for 3D-aware image synthesis largely depend on the 3D pose distribution pre-estimated on the training set. An inaccurate estimation may mislead the model into learning faulty geometry. This work proposes PoF3D that frees generative radiance fields from the requirements of 3D pose priors. We first equip the generator with an efficient pose learner, which is able to infer a pose from a latent code, to approximate the underlying true pose distribution automatically. We then assign the discriminator a task to learn pose distribution under the supervision of the generator and to differentiate real and synthesized images with the predicted pose as the condition. The pose-free generator and the pose-aware discriminator are jointly trained in an adversarial manner. Extensive results on a couple of datasets confirm that the performance of our approach, regarding both image quality and geometry quality, is on par with state of the art. To our best knowledge, PoF3D demonstrates the feasibility of learning high-quality 3D-aware image synthesis without using 3D pose priors for the first time.