Joshua Li

Josh Li

I'm a 4th year computer science undergrad at the University of Waterloo, interested in 3D/4D reconstruction. My long-term goal is to learn a simulated world that is indistinguishable from our "real world". I believe in simple, scalable solutions towards that end.

Currently, I work on feedforward 4D gaussian reconstruction at Waabi. Previously, I was fortunate enough to work with Prof. Yuri Boykov, Prof. Yuhao Chen and Dr. Gregory Schwartz.

More about my personal interests here.

Email / Scholar / Github

Peggy's Cove, NS

Research

	CRF Loss is How Networks Should Learn Boundaries in Weakly Supervised Segmentation Joshua Li, Yuri Boykov arXiv Preprint, 2026 code / arXiv A Conditional Random Field (CRF)-inspired loss, using soft pseudo-labels as unary supervision and high-level boundaries as pairwise supervision. We also show that collision cross-entropy is a robust alternative to standard cross-entropy.
	SHARE: Scene-Human Aligned Reconstruction Joshua Li, Brendan Chharawala, Chang Shu, Xue Bin Peng, Pengcheng Xi SIGGRAPH Asia Tech. Comm., 2025 code / arXiv 3D human motion and scene reconstruction from an RGB video. Uses depth priors to improve placement of human mesh.
	Design Decisions that Matter: Modality, State, and Action Horizon in Imitation Learning Brendan Chharawala, Joshua Li, Stephie Liu, Shawn Yang, Colin Bellinger, David Liu, Chang Shu, Yue Hu, Pengcheng Xi CoRL Workshop, 2025 pdf Effect of teleoperation modality for training generalist robot policy (Octo) on a robotic arm.
	SAMJAM: Zero-Shot Video Scene Graph Generation for Egocentric Kitchen Videos Joshua Li, Fernando Jose Pena Cantu, Emily Yu, Alexander Wong, Yuchen Cui, Yuhao Chen CVPR Workshop, 2025 code / arXiv Video scene graph generation in dynamic kitchen environments. Uses a matching algorithm on Gemini and SAM2 outputs to ensure stable object identities.
	CellNEST reveals cell-cell relay networks using attention mechanisms on spatial transcriptomics Fatema Tuz Zohora, Deisha Paliwal, Eugenia Flores-Figueroa, Joshua Li, Tingxiao Gao, Faiyaz Notta, Gregory W. Schwartz Nature Methods, 2025 vis code / article Graph attention network with contrastive learning to detect cell-cell communication. I wrote code to visualize results.