This is Yucheng Han’s homepage.
I have graduated from Tsinghua University. Now I am going to be a Ph.D. student at Nanyang Technological University advised by Professor Hanwang zhang.
I mainly focus on computer vision. I can be contacted using email yucheng002@e.ntu.edu.sg.
Publication and manuscript
Prompt-aligned Gradient for Prompt Tuning |
|
Fast AdvProp |
|
Robust Process Identification from Step Response Data and Parallel Implementation |
|
Hierarchical Graph with Relation Reduction for Video Summarization |
|
Learning Multiscale Hierarchical Attention for Video Summarization |
Previous Research experiences
Undergraduate Thesis in Department of Automation
Robust Process Identification from Step Response Data and Parallel Implementation
Advised by Professor Chao Shang and Professor Dexian Huang
- In this paper, we propose an improved robust process identification approach from step response data based on Huber loss, which is less sensitive to outliers than generic MSE, and leads to a higher successful rate.
- We show that this solution procedure can be parallelized, which leads to significant computation savings with graphical processing units used, and thus better conforms to requirement in practical situations.
- The corresponding paper is accepted by International Conference on Industrial Artificial Intelligence (IAI 2021).
CCVL Lab
Fast AdvProp
Advised by Professor Cihang Xie and Professor Alan Yuille.
- AdvProp reimplemented using PyTorch is open source.
- Accepted by ICLR 2022.
i-VisonGroup
I have finished two projects in i-VisionGroup and gotten two corresponding papers.
Hierarchical Graph with Relation Reduction for Video Summarization
Advised by Professor Jiwen Lu
- Designed a new video summarization method to extract the spatial-temporal representations by building a
hierarchical graph and applying graph convolution and our proposed pooling module. - Exceeded previous state-of-the-art method by 2.1% on SumMe dataset and 1.6% on TVSum dataset.
Learning Multiscale Hierarchical Attention for Video Summarization
Advised by Professor Jiwen Lu
- Designed a hierarchical attention model for video summarization extracting the multiscale features.
- Benchmarked the proposed attention model on the task of video summarization (as well as a variant of the
attention model that was combined with optical flow) and improved the performance on SumMe and TVSum by
1.4% (1.7%) and 0.4% (0.9%), respectively.