FIELD: computer technology.
SUBSTANCE: present invention relates to a method and a device for creation of a multilayered scene representation. In the method, a pair of scene images is received, in which one image of the pair is considered as a reference image, and the other image of the pair is considered as an initial image; a reprojection operation is performed relatively to the pair of images for creation of a volume expanded on a plane; using a geometrical network Fg, a multilayered scene structure is predicted based on the volume expanded on the plane, and, using a color network Fc, color values and non-transparency values α are assessed for the predicted multilayered scene structure to obtain a multilayered scene representation; wherein, the geometrical network and the color network are trained in an end-to-end way, while the geometrical network is trained using at least geometrical loss calculated between the predicted multilayered structure and a real sparse point cloud corresponding to the scene in the reference image selected from a training video of a training dataset.
EFFECT: increase in the quality of new image synthesis.
16 cl, 14 dwg
Title | Year | Author | Number |
---|---|---|---|
METHOD FOR VISUALIZING A 3D PORTRAIT OF A PERSON WITH ALTERED LIGHTING AND A COMPUTING DEVICE FOR IT | 2021 |
|
RU2757563C1 |
METHOD FOR 3D RECONSTRUCTION OF A HUMAN HEAD TO OBTAIN A RENDER IMAGE OF A PERSON | 2022 |
|
RU2786362C1 |
VISUALIZATION OF RECONSTRUCTION OF 3D SCENE USING SEMANTIC REGULARIZATION OF NORMALS TSDF WHEN TRAINING NEURAL NETWORK | 2023 |
|
RU2825722C1 |
RAPID TWO-LAYER NEURAL NETWORK SYNTHESIS OF REALISTIC IMAGES OF A NEURAL AVATAR BASED ON A SINGLE IMAGE | 2020 |
|
RU2764144C1 |
METHOD OF CREATING FULL-LENGTH ANIMATED AVATAR OF PERSON FROM ONE IMAGE OF PERSON, COMPUTING DEVICE AND MACHINE-READABLE MEDIUM FOR IMPLEMENTATION THEREOF | 2023 |
|
RU2813485C1 |
METHOD FOR CONSTRUCTING A SCENE REPRESENTATION WITH DIRECT CORRECTION FOR REAL-TIME IMAGE SYNTHESIS | 2022 |
|
RU2799237C1 |
METHOD FOR PROVIDING COMPUTER VISION | 2022 |
|
RU2791587C1 |
IMAGE GENERATORS WITH CONDITIONALLY INDEPENDENT PIXEL SYNTHESIS | 2021 |
|
RU2770132C1 |
MODELLING PEOPLE'S CLOTHING BASED ON MULTIPLE POINTS | 2021 |
|
RU2776825C1 |
NEURAL DOT GRAPHIC | 2019 |
|
RU2729166C1 |
Authors
Dates
2023-01-13—Published
2021-07-20—Filed