Jianman Lin

Jianman Lin

Jianman Lin (linjianman). M.Eng. in Electronic Information @ SCUT. Research interests: computer vision, generative AI, controllable image synthesis.

Geometry-Editable and Appearance-Preserving Object Composition

Published in International Journal of Computer Vision (IJCV) — under review · CCF-A, 2025

We propose DGAD, which disentangles object composition into implicit geometric modeling and explicit appearance alignment via dense cross-attention retrieval, improving geometry editing while preserving fine-grained texture.

Recommended citation: Jianman Lin, Haojie Li, Chunmei Qing, Zhijing Yang, Liang Lin, Tianshui Chen. Geometry-Editable and Appearance-Preserving Object Composition. IJCV (submitted). arXiv:2505.20914.
Download Paper