VISION: Video Inpainting for Seamless Instant Object Removal and Background Reconstruction

Video inpainting plays a vital role in video editing and restoration. Removing undesired moving objects poses significant challenges in maintaining background coherence after removing the moving objects. Our proposed work “VISION”, employs a structured four-step approach to achieve this: initially, advanced Natural Language Processing (NLP) techniques are used for semantic understanding. This is followed by leveraging CLIP and Detectron2 for precise target identification and instance segmentation. Subsequently, masks are generated via OpenCV. Finally, moving object removal and stationary background restoration are accomplished through a sophisticated frame restoration technique employing random walk inpainting which is a graph-based algorithm to effectively inpaint masked areas, ensuring high-fidelity results suitable for comprehensive video enhancement and restoration.