Pose synthesis in unseen human poses
AssigneeIntel Corporation
InventorsFanny Nina Paravecino, James Hall, Rita Brugarolas Brufau
Life of this patent
Estimating the term…
Abstract
Techniques related to synthesizing an image of a person in an unseen pose are discussed. Such techniques include detecting a body part occlusion for a body part in a representation of the person in a first image and, in response to the detected occlusion, projecting a representation of the body part from a second image having a different view into the first image. A geometric transformation based on a source pose of the person and a target pose is then applied to the merged image to generate a synthesized image comprising a representation of the person in the target pose.
Claims22
An apparatus for generating an image of a person in a synthesized pose comprising:
- a memory to store pose data corresponding to a person represented in a first image, wherein the first image corresponds to a first view of a scene comprising the person;
- and
- a processor coupled to the memory, the processor to:
- generate a first body part image for the person using at least the first image, wherein the first body part image identifies one of an available set of body parts for pixels of the first image;
- detect a body part occlusion for a first body part of the representation of the person from the first view;
- identify a second image corresponding to a second view of the scene having a second representation of the first body part of the person;
- project the second representation of the first body part in the second image to the first image to generate a merged image;
- and
- apply a geometric transformation to the merged image based at least on the pose data and a target pose to generate a synthesized image comprising a representation of the person in the target pose.
A method for generating an image of a person in a synthesized pose comprising:
- receiving pose data corresponding to a person represented in a first image, wherein the first image corresponds to a first view of a scene comprising the person;
- generating a first body part image for the person using at least the first image, wherein the first body part image identifies one of an available set of body parts for pixels of the first image;
- detecting a body part occlusion for a first body part of the representation of the person from the first view;
- identifying a second image corresponding to a second view of the scene having a second representation of the first body part of the person;
- projecting the second representation of the first body part in the second image to the first image to generate a merged image;
- and
- applying a geometric transformation to the merged image based at least on the pose data and a target pose to generate a synthesized image comprising a representation of the person in the target pose.
At least one non-transitory machine readable medium comprising a plurality of instructions that, in response to being executed on a computing device, cause the computing device to generate an image of a person in a synthesized pose by:
- receiving pose data corresponding to a person represented in a first image, wherein the first image corresponds to a first view of a scene comprising the person;
- generating a first body part image for the person using at least the first image, wherein the first body part image identifies one of an available set of body parts for pixels of the first image;
- detecting a body part occlusion for a first body part of the representation of the person from the first view;
- identifying a second image corresponding to a second view of the scene having a second representation of the first body part of the person;
- projecting the second representation of the first body part in the second image to the first image to generate a merged image;
- and
- applying a geometric transformation to the merged image based at least on the pose data and a target pose to generate a synthesized image comprising a representation of the person in the target pose.
Dependencies are inferred from the claim text.