SMIRK is pioneering the reconstruction of expressive 3D faces with a new training strategy and leveraging expression diversity. By employing a neural rendering technique rather than differentiable rendering, SMIRK can concentrate on facial geometry.

• Neural rendering replaces differentiable rendering for focused geometry training.
• Sampling from training images allows greater expression diversity.
• Superior performance in perceptual evaluations for accurate expression reconstruction.
• Webpage for additional insights: SMIRK Project

This research is groundbreaking in its potential for applications like digital avatars, VR and animation. SMIRK’s capabilities in capturing subtle and extreme expressions promise enhanced realism in digital representations of human emotions.