A User-Guided and Local Motion-Adaptive Framework for Virtual Product Placement in Video

Le résumé fourni par la source

Virtual product placement (VPP) aims to seamlessly insert a product into a pre-recorded video. It is gaining attention as a flexible and scalable strategy for digital advertising, where user preferences and subjective choices should be considered. However, the conventional VPP methods adopt fully automatic insertion frameworks without user involvement, which limits their adaptability to subjective scenarios. To address this limitation, we proposed a user-guided and local motion-adaptive framework, called ULM-VPP, for video VPPs. Given a video and a virtual product, the proposed system obtains an initial target region for the product placement and nearby keypoints in the first frame for minimal user interaction. For each subsequent frame, both the target region and virtual product are realigned based on local motion inferred from the keypoints. Finally, the virtual product is inserted seamlessly into the updated region over all frames while maintaining temporal consistency. Experiments across various video sequences demonstrate that ULM-VPP produces geometrically stable and visually coherent results. The source code is available at https://github.com/halikes/ULM-VPP

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.