Back to results
US 2023/0401716 A1Published application
Transformer for efficient image segmentation
AssigneeAdobe Inc.
InventorsYilin Wang, Chenglin Yang, Jianming Zhang, He Zhang, Zijun Wei, Zhe Lin
Life of this patent
Estimating the term…
Abstract
Systems and methods for image segmentation are described. Embodiments of the present disclosure receive an image depicting an object; generate image features for the image by performing a convolutional self-attention operation that outputs a plurality of attention-weighted values for a convolutional kernel applied at a position of a sliding window on the image; and generate label data that identifies the object based on the image features.
Claims20
A method comprising:
- receiving an image depicting an object;
- generating image features for the image by performing a convolutional self-attention operation that outputs a plurality of attention-weighted values for a convolutional kernel applied at a position of a sliding window on the image;
- and
- generating label data that identifies the object based on the image features.
1 more element
A method comprising:
- receiving training data including a training image and ground-truth label data;
- generating image features for the training image by performing a convolutional self-attention operation using a machine learning model, wherein the machine learning model generates a plurality of attention-weighted values for a convolutional kernel applied at a position of a sliding window on the training image;
- generating label data that identifies an object in the training image based on the image features;
- computing a loss function based on the generated label data and the ground-truth label data;
- and
- updating the machine learning model based on the loss function.
3 more elements
An apparatus comprising:
- an encoder comprising a convolutional self-attention transformer block configured to generate image features for an image by performing a convolutional self-attention operation that outputs a plurality of attention-weighted values for a convolutional kernel applied at a position of a sliding window on the image;
- and
- a decoder configured to generate label data that identifies an object of the image based on the image features.
Dependencies are inferred from the claim text.