Publications

First Extensive Survey: We present the first comprehensive survey of semantic correspondence, proposing a taxonomy to classify existing semantic correspondence methods and providing a detailed analysis of their designs.

Unified Benchmarking: We aggregate results across various benchmarks into a unified comparative table and conduct thorough controlled experiments to analyze the effectiveness of different method components.

Strong Baseline: We propose a simple yet effective baseline that achieves state-of-the-art performance on multiple benchmarks, providing a solid foundation for future research.

Feed-forward 3DGS Framework: We introduce a novel framework leveraging 3D Gaussian Splatting for high-fidelity, animatable avatar reconstruction in a single forward pass, bypassing the need for per-subject optimization.

Feature Fusion Transformer: We design a deformable attention transformer to efficiently fuse 3D representation features with semantic image features, ensuring geometric and appearance fidelity in the canonical space.

Adaptive Cloth Joint Learning: We propose an adaptive strategy extending Linear Blend Skinning (LBS) beyond the SMPL template to model complex dynamic clothing deformations realistically across diverse styles.