Back to results
US 11,380,119 B2Granted
Pose-aligned networks for deep attribute modeling
AssigneeMeta Platforms, Inc.
InventorLubomir Dimitrov Bourdev
Life of this patent
Estimating the term…
Abstract
In one embodiment, a method includes locating a plurality of part patches from an image, wherein each part patch comprises at least a portion of the image corresponding to a recognized human body portion or pose, and wherein each part patch is associated with a respective detection score larger than a threshold score, wherein the detection score is determined based on a comparison between the part patch with multiple training patches, generating a plurality of sets of feature data by processing each of the plurality of part patches with a plurality of convolutional neural networks, respectively, and determining whether a human attribute exists in the image based on the plurality of sets of feature data.
Claims19
A method comprising, by a computing device:
- locating a plurality of candidate patches from an image, wherein each candidate patch comprises at least a portion of the image;
- determining a plurality of part patches for the image based on comparisons between each candidate patch with multiple training patches corresponding to multiple distinct human body portions or poses, wherein each determined part patch comprises a candidate patch having a closeness greater than a threshold value to a training patch corresponding to one of the multiple human body portions or poses, and wherein the plurality of determined part patches correspond to one or more of the multiple distinct human body portions or poses;
- generating a plurality of sets of feature data corresponding to the one or more distinct human body portions or poses by processing each of the plurality of part patches with a plurality of first convolutional neural networks, respectively;
- generating a set of whole-person feature data corresponding to a whole-person region covering an entire human body depicted in the image by processing the whole-person region in the image with a second convolutional neural network;
- concatenating the plurality of sets of feature data corresponding to the one or more distinct human body portions or poses to generate a first set of feature data;
- incorporating the set of whole-person feature data into the first set of feature data to generate a second set of feature data;
- and
- determining, based on the second set of feature data, whether a human attribute exists in the image.
5 more elements
One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
- locate a plurality of candidate patches from an image, wherein each candidate patch comprises at least a portion of the image;
- determine a plurality of part patches for the image based on comparisons between each candidate patch with multiple training patches corresponding to multiple distinct human body portions or poses, wherein each determined part patch comprises a candidate patch having a closeness greater than a threshold value to a training patch corresponding to one of the multiple human body portions or poses, and wherein the plurality of determined part patches correspond to one or more of the multiple distinct human body portions or poses;
- generate a plurality of sets of feature data corresponding to the one or more distinct human body portions or poses by processing each of the plurality of part patches with a plurality of first convolutional neural networks, respectively;
- generate a set of whole-person feature data corresponding to a whole-person region covering an entire human body depicted in the image by processing the whole-person region in the image with a second convolutional neural network;
- concatenate the plurality of sets of feature data corresponding to the one or more distinct human body portions or poses to generate a first set of feature data;
- incorporate the set of whole-person feature data into the first set of feature data to generate a second set of feature data;
- and
- determine, based on the second set of feature data, whether a human attribute exists in the image.
5 more elements
A system comprising:
- one or more processors;
- and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
- locate a plurality of candidate patches from an image, wherein each candidate patch comprises at least a portion of the image;
- determine a plurality of part patches for the image based on comparisons between each candidate patch with multiple training patches corresponding to multiple distinct human body portions or poses, wherein each determined part patch comprises a candidate patch having a closeness greater than a threshold value to a training patch corresponding to one of the multiple human body portions or poses, and wherein the plurality of determined part patches correspond to one or more of the multiple distinct human body portions or poses;
- generate a plurality of sets of feature data corresponding to the one or more distinct human body portions or poses by processing each of the plurality of part patches with a plurality of first convolutional neural networks, respectively;
- generate a set of whole-person feature data corresponding to a whole-person region covering an entire human body depicted in the image by processing the whole-person region in the image with a second convolutional neural network;
- concatenate the plurality of sets of feature data corresponding to the one or more distinct human body portions or poses to generate a first set of feature data;
- incorporate the set of whole-person feature data into the first set of feature data to generate a second set of feature data;
- and
- determine, based on the second set of feature data, whether a human attribute exists in the image.
7 more elements
Dependencies are inferred from the claim text.