Lihan Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Lihan Zhang has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Aoyang Liu
Shenzhen International Graduate School, Tsinghua University, Shenzhen, China
1 shared publication - Chi Zhang
Institute of Artificial Intelligence (TeleAI), China Telecom, Shanghai, China
1 shared publication - Haibin Huang
Institute of Artificial Intelligence (TeleAI), China Telecom, Shanghai, China
1 shared publication - Wei-Shi Zheng
Sun Yat-sen University
1 shared publication - Xiaojun Liang
Peng Cheng Laboratory, Shenzhen, China
1 shared publication - Xiaoyan Yang
Institute of Artificial Intelligence (TeleAI), China Telecom, Shanghai, China
1 shared publication
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 16085-16094
Click2Insert: Video Object Insertion via Sparse Click Guidance
With the rapid progress of video editing technologies, video object insertion has attracted increasing attention. Existing approaches typically rely on either text instructions, which often lack precise spatial control, or dense masks, which are costly and impractical to annotate manually. To address these limitations, we present Click2Insert, a video object insertion framework that enables precise insertion using only a few point annotations on keyframes. Our framework is also compatible with mask inputs, enabling flexible control and supporting mask-based editing across different scenarios. To train the model, we further construct a large-scale paired insertion dataset containing 1.2M samples through an automated pipeline. Furthermore, we improve sparse-guided video object insertion from three perspectives: condition compatibility, spatial accuracy, and temporal coherence. Specifically, we introduce Task Aware Modulation to support both point- and mask-based conditions within a single model, Spatial-Aware Loss to improve positional accuracy under sparse point supervision, and Intra-Chunk Temporal Attention to enhance temporal consistency across adjacent frames. Extensive experiments demonstrate that our method achieves superior editing performance, enabling precise object insertion in a more user-friendly and flexible manner.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 16085-16094
Open AccessClick2Insert: Video Object Insertion via Sparse Click Guidance
With the rapid progress of video editing technologies, video object insertion has attracted increasing attention. Existing approaches typically rely on either text instructions, which often lack precise spatial control, or dense masks, which are costly and impractical to annotate manually. To address these limitations, we present Click2Insert, a video object insertion framework that enables precise insertion using only a few point annotations on keyframes. Our framework is also compatible with mask inputs, enabling flexible control and supporting mask-based editing across different scenarios. To train the model, we further construct a large-scale paired insertion dataset containing 1.2M samples through an automated pipeline. Furthermore, we improve sparse-guided video object insertion from three perspectives: condition compatibility, spatial accuracy, and temporal coherence. Specifically, we introduce Task Aware Modulation to support both point- and mask-based conditions within a single model, Spatial-Aware Loss to improve positional accuracy under sparse point supervision, and Intra-Chunk Temporal Attention to enhance temporal consistency across adjacent frames. Extensive experiments demonstrate that our method achieves superior editing performance, enabling precise object insertion in a more user-friendly and flexible manner.
- Open Access
- 12 January 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (1) , 1-24
Enhancing Pose-Guided Human Image Generation with Comprehensive and Adjustable 3D Control
Pose-guided human image generation aims to render a source image in a specific pose. Current methods predominantly employ 2D-based signals, which exhibit inherent information deficits, as pose conditions. This leads to difficulty in establishing precise source-target appearance-pose correspondence and further causing uncertainty in predicting self-occluded regions’ appearance. To address these issues, we propose a 3D Pose Conditional Diffusion model (3DPCD) that leverages a human parametric model to integrate comprehensive and adjustable 3D control into forward–backward diffusion steps. Specifically, we employ Fourier-transformed SMPL-X as the 3D pose representation to facilitate precise source-target correspondence by understanding the complete pose information. Building on this, we further propose a hierarchical appearance-pose alignment method, which aligns appearance with the complete pose information at both global and local levels. Moreover, motivated by the fact that human pose transformation is a progressive process in 3D space and our 3D pose representation is adjustable, we integrate progressively interpolated 3D control into a series of sampling steps. This effectively mitigates uncertainties in pixel transfer between poses. It should be noted that the proposed explicit pose-guided strategy also supports flexible adjustment of pose, shape, and viewpoint. Both quantitative and qualitative evaluations demonstrate that our 3DPCD outperforms state-of-the-art methods on the widely used DeepFashion InShop benchmark and our newly constructed PoseWeb-33 dataset, which features richer appearance variations and more diverse conditional poses.
- 12 January 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (1) , 1-24
Open AccessEnhancing Pose-Guided Human Image Generation with Comprehensive and Adjustable 3D Control
Pose-guided human image generation aims to render a source image in a specific pose. Current methods predominantly employ 2D-based signals, which exhibit inherent information deficits, as pose conditions. This leads to difficulty in establishing precise source-target appearance-pose correspondence and further causing uncertainty in predicting self-occluded regions’ appearance. To address these issues, we propose a 3D Pose Conditional Diffusion model (3DPCD) that leverages a human parametric model to integrate comprehensive and adjustable 3D control into forward–backward diffusion steps. Specifically, we employ Fourier-transformed SMPL-X as the 3D pose representation to facilitate precise source-target correspondence by understanding the complete pose information. Building on this, we further propose a hierarchical appearance-pose alignment method, which aligns appearance with the complete pose information at both global and local levels. Moreover, motivated by the fact that human pose transformation is a progressive process in 3D space and our 3D pose representation is adjustable, we integrate progressively interpolated 3D control into a series of sampling steps. This effectively mitigates uncertainties in pixel transfer between poses. It should be noted that the proposed explicit pose-guided strategy also supports flexible adjustment of pose, shape, and viewpoint. Both quantitative and qualitative evaluations demonstrate that our 3DPCD outperforms state-of-the-art methods on the widely used DeepFashion InShop benchmark and our newly constructed PoseWeb-33 dataset, which features richer appearance variations and more diverse conditional poses.