Transformer-Guided Video Inpainting Algorithm Based on Local Spatial-Temporal joint

INTRODUCTION: Video inpainting is a very important task in computer vision, and it’s key component of various practical applications. It also plays an role video occlusion removal, traffic monitoring old movie restoration technology. to obtain reasonable content from the sequence fill missing region, maintain time continuity spatial consistency.OBJECTIVES: In previous studies, due complexity scene inpainting, there are often cases fast motion objects or background objects, which will lead optical flow failure. So current algorithm hasn’t met requirements order avoid problem failure, this paper proposes transformer-guided model based on local Spatial-temporal joint.METHODS: First, considering rich relationship between flows, Local Spatial-Temporal Joint Network (LSTN) including encoder, decoder transformer module designed roughly inpaint corrupted frames, Deep Flow used calculate bidirectional flows. Then, map input into Completion (LFCN) with pseudo 3D convolution attention mechanism complete set maps. Finally, inpainted frame sent output.RESULTS: Experiments show that achieves high quality results target removal task, has certain improvement indicators compared advanced technologies.CONCLUSION: Transformer-Guided Inpainting Algorithm Based joint can high-quality information result video.