B‑spline–based method for 2D animation inbetweening that preserves outline clarity and handles large motion spans; combines contour vectorization, structural matching, clustering, and quadratic trajectory prediction.
Complete reproduction and critical analysis of Simo-Serra et al.’s rough sketch simplification paper; includes Lua/Torch to Python/PyTorch migration, loss function validation, and dataset construction experiments.
Undergraduate thesis proposing a dual-LightGCN model to decompose implicit feedback into Preference and Exposure matrices, substantially reducing estimation variance and improving performance on benchmark datasets (Yahoo!R3, Coat).
Submitted to CVPR 2026
This paper is about ManiTaskGen, a universal system that generates a comprehensive set of feasible mobile manipulation tasks given arbitrary scenes. These tasks facilitate both benchmarking and the improvement of embodied decision-making agents.
Recommended citation: Liu Dai* ,Haina Wang*, Weikang Wan, and Hao Su. (2025). "ManiTaskGen: A Comprehensive Task Generator for Benchmarking and Improving Vision-Language Agents on Embodied Decision-Making." arXiv preprint arXiv:2505.20726.
Download Paper
Undergraduate Teaching Assistant(UTA), Zhejiang University, School of Economics and Management, 2022