Region-Adaptive Texture Enhancement for Detailed Person Image Synthesis

The ability to produce convincing textural details is essential
for the fidelity of synthesized person images. Existing meth-
ods typically follow a “warping-based” strategy that propa-
gates appearance features through the same pathway used for
pose transfer. However, most fine-grained features would be
lost during down-sampling, leading to over-smoothed clothes
and missing details in the output images. In this paper we
presents RATE-Net, a novel framework for synthesizing per-
son images with sharp texture details. The proposed frame-
work leverages an additional texture enhancing module to ex-
tract appearance information from the source image and esti-
mate a fine-grained residual texture map, which helps to re-
fine the coarse estimation from the pose transfer module. In
addition, we design an effective alternate updating strategy
to promote mutual guidance between two modules for better
shapeandappearanceconsistency. Experimentsconductedon
DeepFashion benchmark dataset have demonstrated the supe-
riority of our framework compared with existing networks.