Egor Zakharov

I am a Postdoc at ETH Zurich AIT lab. My research interests include computer vision and graphics for the problems of 3D reconstruction and image animation.

I received my Ph.D. from Skoltech, under the supervision of Victor Lempitsky. During my Ph.D., I also was a Lead Research Scientist at Samsung AI Center.

Email  /  Google Scholar  /  Twitter

Representative papers are highlighted.

Highly expressive and animatable 3D human head avatars for VR telepresence.

Reconstruction of highly personalized and realistically rendered strand-based hairstyles from a monocular video.

Instant and animatable 3D human head avatars from a single image with disentangled expression reenactment.

Text-guided diffusion model for strand-based hairstyles generation & editing.

Reconstructing personalized and realistic hairstyles in the form of 3D strands from a smartphone video.

We improve existing surface reconstruction methods using a plug-and-play hybrid representation, achieving up-to 18% improvement in reconstruction quality and approx. 2x faster convergence speed.

We create animatable 3D head reconstructions in the form of a rigged mesh from a single image by leveraging facial priors and differentiable rendering of in-the-wild video data.

Learning disentangled expression and appearance representations for high-resolution portrait animation.

Speeding-up the inference of neural head avatars using an efficient architecture that achieves real-time performance on mobile.

Leveraging a large corpus of weakly annotated video data to achieve the realistic synthesis of human heads in a few-shot scenario using meta-learning and GAN-based training.

We integrated classical and neural rendering approaches to achieve high robustness of the avatar renders to the poses and views unseen during training.

We combine the GAN-based and perceptual losses in a single training method, which primarily excels at the task of facial attribute manipulation.

This website uses a template by Jon Barron