Surface normal prediction using pair-wise angular training
AssigneeAdobe Inc.
InventorJianming ZHANG
Life of this patent
Estimating the term…
Abstract
A surface normal model is trained to predict normal maps from single images using pair-wise angular losses. A training dataset comprising a training image and a ground truth normal map for the training image is received. To train the surface normal model using the training dataset, a predicted normal map is generated for the training image using the surface normal model. A loss is determined as a function of angular values between pairs of normal vectors for the predicted normal map and corresponding angular values between pairs of normal vectors for the ground truth normal map. The surface normal model is updated based on the loss.
Claims20
One or more computer storage media storing computer-useable instructions that, when used by a computing device, cause the computing device to perform operations, the operations comprising:
- receiving a training dataset comprising a training image and a ground truth normal map for the training image;
- and
- training a surface normal model using the training dataset by:
- generating, by the surface normal model, a predicted normal map using the training image;
- determining a loss as a function of angular values between pairs of normal vectors for the predicted normal map and corresponding angular values between pairs of normal vectors for the ground truth normal map;
- and
- updating the surface normal model based on the loss.
A computer-implemented method comprising:
- receiving, by a surface normal training component, a training dataset comprising a training image and a ground truth normal map for the training image;
- and
- training, by the surface normal training component, a surface normal model using a loss determined as a function of differences between angular values for pairs of normal vectors for a predicted normal map generated from the training image by the surface normal model and corresponding angular values between pairs of normal vectors for the ground truth normal map.
A computer system comprising:
- one or more processors;
- and
- one or more computer storage media storing computer-useable instructions that, when used by the one or more processors, causes the one or more processors to perform operations comprising:
- receiving, by a surface normal prediction component, an input image;
- and
- generating, by the surface normal prediction component, a normal map from the input image using a surface normal model trained using a loss determined as a function of differences between angular values for pairs of normal vectors for a predicted normal map generated by the surface normal map using a training image by the surface normal model and corresponding angular values between pairs of normal vectors for a ground truth normal map for the training image.
Dependencies are inferred from the claim text.