Pareto Optimized Large Mask Approach for Efficient and Background Humanoid Shape Removal

We introduce deep learning-based methodology for removing unwanted human-like shapes in videos. The method uses Pareto-optimized Generative Adversarial Networks (GANs) technology, which is a novel contribution. system automatically selects the Region of Interest (ROI) each humanoid shape and skeleton detection module to determine retain. semantic masks human like are created using semantic-aware occlusion-robust model that has four primary components: feature extraction, local, global, branches. global branch encodes occlusion-aware information make extracted features resistant occlusion, while local retrieves fine-grained characteristics. A modified big mask inpainting approach employed eliminate person from image, leveraging Fast Fourier convolutions utilizing polygonal chains rectangles with unpredictable aspect ratios. inpainter network takes input image create an output excluding background shapes. generator encoder-decoder structure included skip connections recover spatial dilated convolution squeeze excitation blocks regions behind consistent their surroundings. discriminator avoids dissimilar at patch scale, refiner catches around boundaries shape. evaluated on two video object segmentation datasets (DAVIS YouTube-VOS) database 66 distinct sequences people desk office environment. efficiency was assessed Structural Similarity Index Measure (SSIM), Frechet Inception Distance (FID), Learned Perceptual Image Patch (LPIPS) metrics showed promising results fully automated removal task.