US 12,475,163 B2
Method and system for a text-vision retrieval framework
Min Gong, Shanghai (CN); Qi Bao, Acton, MA (US); and Zijia Wang, London (GB)
Assigned to DELL PRODUCTS L.P., Round Rock, TX (US)
Filed by Dell Products L.P., Round Rock, TX (US)
Filed on Mar. 25, 2024, as Appl. No. 18/615,042.
Prior Publication US 2025/0298840 A1, Sep. 25, 2025
Int. Cl. G06F 16/53 (2019.01); G06F 16/532 (2019.01); G06F 16/535 (2019.01); G06F 16/538 (2019.01); G06F 16/56 (2019.01)
CPC G06F 16/535 (2019.01) [G06F 16/532 (2019.01); G06F 16/538 (2019.01); G06F 16/56 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method for managing an image retrieval framework, the method comprising:
receiving, by a graphical user interface (GUI), a query about an object from a user, wherein the GUI sends the query to an intention recognition module (IRM);
analyzing, by the IRM, the query to infer intention of the user;
in response to the analyzing, making, by the IRM, a determination that the intention is searching for the object in a semantic text database, wherein, based on the determination, the IRM sends the query to a catalog search module (CSM);
transforming, by the CSM and via a single modality embedding transform model (SMETM), the query into a first embedding vector;
performing, by the CSM and using the first embedding vector, a search for a most matching object in a semantic text database;
identifying, by the CSM and in response to the search, a path associated with the query;
fetching, by the CSM and based on the path, a reference image (RI) from a catalog image store, wherein the RI is sent to the GUI to obtain a confirmation from the user;
sending, by the CSM and after obtaining the confirmation, the RI to an image embedding generation module (IEGM);
transforming, by the IEGM and via a multimodal embedding transform model (METM), the RI into a second embedding vector, wherein the IEGM sends the second embedding vector to an image by image module (IIM);
performing, by the IIM and using the second embedding vector, a second search for a nearest image to the RI in a semantic image database;
identifying, by the IIM and in response to the second search, a second path associated with the nearest image, wherein the IIM sends the second path to an analyzer, wherein the analyzer further sends the second path to an image access module (IAM);
fetching, by the IAM and based on the second path, an image from a retrieval image store, wherein the IAM sends the image to the GUI; and
initiating, by the GUI, displaying of the image to the user in response to the query.