Coupled multi-task fully convolutional networks using multi-scale contextual information and hierarchical hyper-features for semantic image segmentation
AssigneeIntel Corporation
InventorsLibin Wang, Anbang Yao, Yurong Chen
Life of this patent
Estimating the term…
Abstract
Techniques related to implementing fully convolutional networks for semantic image segmentation are discussed. Such techniques may include combining feature maps from multiple stages of a multi-stage fully convolutional network to generate a hyper-feature corresponding to an input image, up-sampling the hyper-feature and summing it with a feature map of a previous stage to provide a final set of features, and classifying the final set of features to provide semantic image segmentation of the input image.
Claims20
A system comprising:
- a memory to store at least a portion of an input image;
- and
- processor circuitry coupled to the memory, the processor circuitry to:
- implement a multi-stage convolutional network based on the input image via application of first, second, and third stages to output a first feature map at a first resolution, a second feature map at a second resolution less than the first resolution, and a third feature map at the second resolution, respectively;
- combine at least the second and third feature maps to generate a hyper-feature corresponding to the input image;
- up-sample the hyper-feature to the first resolution;
- generate a final set of features via combination of the up-sampled hyper-feature and the first feature map;
- and
- provide a semantic image segmentation of the input image based on the final set of features.
A method comprising:
- implementing a multi-stage convolutional network based on an input image by applying first, second, and third stages to output a first feature map at a first resolution, a second feature map at a second resolution less than the first resolution, and a third feature map at the second resolution, respectively;
- combining at least the second and third feature maps to generate a hyper-feature corresponding to the input image;
- up-sampling the hyper-feature to the first resolution;
- generating a final set of features by combining the up-sampled hyper-feature and the first feature map;
- and
- providing a semantic image segmentation of the input image based on the final set of features.
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 perform semantic image segmentation by:
- implementing a multi-stage convolutional network based on an input image by applying first, second, and third stages to output a first feature map at a first resolution, a second feature map at a second resolution less than the first resolution, and a third feature map at the second resolution, respectively;
- combining at least the second and third feature maps to generate a hyper-feature corresponding to the input image;
- up-sampling the hyper-feature to the first resolution;
- generating a final set of features by combining the up-sampled hyper-feature and the first feature map;
- and
- providing a semantic image segmentation of the input image based on the final set of features.
Dependencies are inferred from the claim text.