US 12,450,871 B2
Enabling feature importance using Siamese autoencoders for effective image change detection
Satish Kumar Mopur, Bangalore (IN); Gunalan Perumal Vijayan, Bangalore (IN); Shounak Bandopadhyay, Bangalore (IN); and Krishnaprasad Lingadahalli Shastry, Bangalore (IN)
Assigned to Hewlett Packard Enterprise Development LP, Spring, TX (US)
Filed by HEWLETT PACKARD ENTERPRISE DEVELOPMENT LP, Houston, TX (US)
Filed on Mar. 29, 2022, as Appl. No. 17/707,612.
Prior Publication US 2023/0316710 A1, Oct. 5, 2023
Int. Cl. G06V 10/762 (2022.01); G06N 3/045 (2023.01); G06V 10/776 (2022.01); G06V 10/82 (2022.01)
CPC G06V 10/763 (2022.01) [G06N 3/045 (2023.01); G06V 10/776 (2022.01); G06V 10/82 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computing device comprising:
a memory; and
one or more processors that are configured to execute machine readable instructions stored in the memory for performing a method comprising:
initiating a training process of a Siamese AutoEncoder, wherein the Siamese AutoEncoder detects dissimilarities between a first pair of images, and wherein the Siamese AutoEncoder comprises two sub neural networks that are each configured to:
generate a feature vector on the first pair of images to create two feature vectors of the first pair of images, and
compare the two feature vectors to detect the dissimilarities;
during the training process:
receiving a second pair of images;
providing the second pair of images to an encoder and a decoder of the Siamese AutoEncoder to generate a decoded second pair of images;
initiating fine tuning of a first loss function and a second loss function with the decoded second pair of images, wherein the first loss function is an adaptive margin loss function that includes an upper margin value and a lower margin value and the second loss function is a contrastive loss function that uses the decoded second pair of images that are dissimilar within a margin value;
determining a similarity value associated with the decoded second pair of images using the first loss function and the second loss function; and
generating an output of image features based on the similarity value.