View source for 18367034. DENSE FEATURE SCALE DETECTION FOR IMAGE MATCHING simplified abstract (Snap Inc.)

=DENSE FEATURE SCALE DETECTION FOR IMAGE MATCHING=


==Organization Name==

[[:Category:Snap Inc.|Snap Inc.]]


[[Category:Snap Inc.]]

==Inventor(s)==

[[:Category:Shenlong Wang of Toronto (CA)|Shenlong Wang of Toronto (CA)]][[Category:Shenlong Wang of Toronto (CA)]]

[[:Category:Linjie Luo of Los Angeles CA (US)|Linjie Luo of Los Angeles CA (US)]][[Category:Linjie Luo of Los Angeles CA (US)]]

[[:Category:Ning Zhang of Los Angeles CA (US)|Ning Zhang of Los Angeles CA (US)]][[Category:Ning Zhang of Los Angeles CA (US)]]

[[:Category:Jia Li of Marina Del Rey CA (US)|Jia Li of Marina Del Rey CA (US)]][[Category:Jia Li of Marina Del Rey CA (US)]]

==DENSE FEATURE SCALE DETECTION FOR IMAGE MATCHING - A simplified explanation of the abstract==


This abstract first appeared for US patent application 18367034 titled 'DENSE FEATURE SCALE DETECTION FOR IMAGE MATCHING

==Simplified Explanation==

The patent application describes a method for detecting dense features in images using multiple convolutional neural networks trained on scale data. This allows for more accurate and efficient matching of pixels between images.

* Multiple convolutional neural networks are trained on scale data to detect dense features in images.
* An input image is used to generate multiple scaled images.
* The scaled images are input into a feature net, which outputs feature data for each scaled image.
* An attention net generates an attention map from the input image, assigning emphasis to different scales based on texture analysis.
* The feature data and attention data are combined through a multiplication process and then summed to generate dense features for comparison.

== Potential Applications ==
* Image recognition and classification
* Object detection and tracking
* Image matching and alignment
* Augmented reality applications

== Problems Solved ==
* Inaccurate and inefficient pixel matching between images
* Difficulty in detecting dense features in images
* Lack of emphasis on different scales based on texture analysis

== Benefits ==
* More accurate and efficient matching of pixels between images
* Improved detection of dense features in images
* Enhanced emphasis on different scales based on texture analysis


==Original Abstract Submitted==

Dense feature scale detection can be implemented using multiple convolutional neural networks trained on scale data to more accurately and efficiently match pixels between images. An input image can be used to generate multiple scaled images. The multiple scaled images are input into a feature net, which outputs feature data for the multiple scaled images. An attention net is used to generate an attention map from the input image. The attention map assigns emphasis as a soft distribution to different scales based on texture analysis. The feature data and the attention data can be combined through a multiplication process and then summed to generate dense features for comparison.



[[Category:G06T7/246]]
[[Category:G06T7/33]]
[[Category:G06V10/764]]
[[Category:G06V10/82]]
[[Category:G06V10/44]]