In the era of hybrid working mode, more and more people prefer online video meetings to face-to-face meetings. Considering that people may have online meetings from home, there are many solutions such as background blur or virtual background to protect user privacy. However, if a user tries to demonstrate certain object when the background blur/virtual background enabled, the object may be viewed as a part of background and then loss the visibility as the picture below. This invention disclosure proposes the idea of utilizing one-shot learning to allow users to specify the objects they want to be visible in a background blur/virtual background enabled video conferencing, along with implementing a post-processing to make any of one-shot learning models perform better for object tracking in video conferencing user scenario. In this disclosure, the user scenario we use to explain the idea is set to show an object in a background blur/virtual background enabled video conferencing. But the application of this idea is not limited to this scenario, for example, it can be also used for auto framing on any of objects users specify instead of only on human faces. Besides, even though this scenario combines image segmentation, gesture detection, and one-shot learning deep learning models, we will only discuss more details on one-shot learning model in this disclosure because it’s the most critical part for the invention, and the rest of the two models are very common in existing applications.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.
INC, HP, "APPLICATION OF ONE-SHOT LEARNING DETECTOR TO AVOID UNWANTED OBJECTS TO BE BLURRED/HIDDEN IN VIDEO CONFERENCING", Technical Disclosure Commons, (November 21, 2022)