| CPC G06V 10/7715 (2022.01) [G06F 16/51 (2019.01); G06F 16/532 (2019.01); G06V 10/758 (2022.01); G06V 10/7753 (2022.01); G06V 10/82 (2022.01); G06V 20/10 (2022.01)] | 23 Claims |
|
1. A computer-implemented method comprising:
applying a query image and a search image to a neural network of a feature extraction network of a computing device, the query image indicating an object to be searched for in the search image, wherein the feature extraction network includes the neural network, a spatial feature neural network coupled to receive a first output of the neural network pertaining to the search image, and an embedding network coupled to receive a second output of the neural network pertaining to the query image;
generating spatial search features from the spatial feature neural network;
indexing the spatial search features from the spatial feature neural network by slicing the search image into sets of spatial features, spatially pooling the sets of spatial features into super-pixels by grouping neighboring features if they exceed a specified similarity threshold, and storing the super-pixels in a database;
generating a query feature from the embedding network;
applying the query feature to an approximate nearest neighbor (ANN) retrieval index; and
building the ANN retrieval index using the super-pixels;
determining cosine similarity between the query feature and the spatial search features to produce a heat map of likely positions of objects in the search image which are similar to the query image; and
determining an optimal matching result of an object in the search image to the query image based on an operation using the ANN retrieval index.
|