Pix2Pix-Hybrid: Structure-Guided Conditional Synthesis of Hajj Crowd Images with Multi-Channel Conditioning and Weak Attribute Supervision

The paper introduces Pix2Pix-Hybrid (P2P-H), a hybrid conditional GAN designed for synthesizing Hajj crowd images to enhance crowd-counting models, addressing data scarcity and privacy issues. P2P-H utilizes a U-Net generator with eight input channels for structural and contextual conditioning, and employs multi-scale PatchGAN discriminators for improved texture detail. The model, trained on 993 real images, resulted in a synthetic dataset of 10,000 high-resolution images and demonstrated significant improvements in synthesis quality and downstream performance, particularly reducing mean absolute error (MAE) in crowd-counting tasks when combined with real data.