Abstract
In many use cases, it is useful to visualize a product within a target environment. This disclosure describes techniques to train and employ image generation artificial intelligence models to generate photorealistic images of one or more products placed within a target physical environment. The user can interact with the generated image to obtain additional views with different placement and/or orientation of the product in different rooms/ room types. The generated images provide the user with a high-quality visualization that enables them to evaluate the fit and aesthetics of the product within the target physical environment. The models are trained to generate images of a room as it would look when the selected product is placed within it. Background masking, inpainting/outpainting, etc. are used to ensure that the generated image fits the room. Model training is performed with a custom training dataset that includes product images from various angles and at various zoom levels, in various contexts, and also includes counterfactual examples.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Gorantla, Siva Kumar; Driscoll, Brendan; Malhi, Ishaan; Pruthi, Garima; Holden, Krista; Dutta, Praneet; and Narayanaswamy, Arunachalam, "Generating Photorealistic Images of Physical Products Placed Within a Target Space", Technical Disclosure Commons, (September 17, 2024)
https://www.tdcommons.org/dpubs_series/7363