HifiFace: 3D Shape and Semantic Prior Guided High Fidelity Face Swapping

Deal with swapping is the task of producing images with resource encounter identification and the attributes from the focus on image. Current approaches encounter complications in supplying photo-reasonable effects or conserving the initial encounter shape.

Deal with recognition – creative interpretation in Hollywood CA. Impression credit history: YO! What Took place To Peace? by means of Flickr, CC BY-SA two.

A current examine proposes a novel finish-to-finish discovering framework, which can maintain the encounter shape and crank out substantial fidelity encounter-swapping effects. A 3D shape-mindful identification extractor generates an identification vector with exact shape information and facts to enforce exact encounter shape transfer. A semantic facial fusion module is proposed to realize a much better mix in function-degree and image-degree. That helps to address occlusion and lights issues.

Substantial experiments exhibit that the advised technique can crank out increased fidelity effects than prior approaches. It can be applied in the film industry, laptop or computer games, or encounter forgery detection.

In this perform, we propose a substantial fidelity encounter swapping technique, referred to as HifiFace, which can perfectly maintain the encounter shape of the resource encounter and crank out photo-reasonable effects. In contrast to other current encounter swapping performs that only use encounter recognition product to maintain the identification similarity, we propose 3D shape-mindful identification to regulate the encounter shape with the geometric supervision from 3DMM and 3D encounter reconstruction technique. Meanwhile, we introduce the Semantic Facial Fusion module to enhance the mix of encoder and decoder characteristics and make adaptive mixing, which can make the effects more photo-reasonable. Substantial experiments on faces in the wild exhibit that our technique can maintain much better identification, especially on the encounter shape, and can crank out more photo-reasonable effects than prior condition-of-the-art solutions.

Study paper: Wang, Y., “HifiFace: 3D Form and Semantic Prior Guided Large Fidelity Deal with Swapping”, 2021 . Backlink to the post: https://arxiv.org/abdominal muscles/2106.09965


Rosa G. Rose

Next Post

Multi-Task Learning and Adapted Knowledge Models for Emotion-Cause Extraction

Wed Jun 23 , 2021
Emotion recognition in the textual content has gotten major notice from researchers lately. Even so, recent types battle with high-quality-grained emotion examination, for occasion, tagging emotion trigger in textual content. As emotion recognition is carefully similar to trigger detection, a the latest examine proposes a framework for modeling them jointly. […]