PERFORMANCE ANALYSIS OF STATE-OF-THE-ART MODELS FOR POSE-GUIDED PERSON IMAGE GENERATION
Biponjot Kaur · INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING & TECHNOLOGY · 2025
Pose-guided person image generation is now a central field of study in computer vision, where sophisticated deep-learning methods are used to generate realistic images of people in a given pose.This work compares the performance of current state-of-theart models on two benchmark datasets: DeepFashion and Market-1501.These datasets provide dense pose, clothing, and background variations and therefore are appropriate for quantifying model robustness.Evaluation is focused on key metrics such as Structural Similarity Index (SSIM), Fréchet Inception Distance (FID), and Inception Score (IS) to estimate the quality, realism, and diversity of the generated images.Our results identify the strengths and weaknesses of each model, providing important insights for future development in pose-guided image synthesis.We also bring into focus the challenges presented by human deformation and structural alignment, which are still the areas of utmost need for improvement.