Quantum Based High Resolution Texture Generation For 3D Human Reconstruction From A Single Image

Abstract

Single-image 3D human reconstruction has used rapidly in geometry estimation, yet photo- realistic texture recovery remains tedious because most of the body surface is unobserved, clothing contains high-frequency patterns, and small pose errors create visible seams in UV space. This research presents QTex-HR, a quantum inspired texture generation framework for reconstructing high resolution human appearance from one RGB image. The method combines an image-conditioned implicit body prior with a UV-domain texture generator whose attention blocks are modulated by parameterized quantum feature maps. Instead of using quantum hardware, QTex-HR simulates quantum encoding principles superposition, phase rotation, and entanglement-like channel mixing inside differentiable neural layers. Sharper hallucinations of occluded backside clothes, hair, and limb textures are made possible by these layers, which strengthen long-range association between symmetric and semantically linked body parts. We evaluate the approach on a controlled synthetic benchmark and a real-image validation set derived from clothed human scans. The proposed model improves texture PSNR by 1.7 dB over a transformer baseline, reduces LPIPS from 0.132 to 0.096, and decreases UV seam error by 42%. Ablation experiments show that quantum mixing is most beneficial when paired with geometry-aware normal conditioning and multi-scale adversarial supervision. The results indicate that quantum-inspired representation learning can be a practical design principle for high-resolution neural texture synthesis without requiring near-term quantum processors.