Hello there!
Welcome to Kunjun Li’s website! I’m currently a third-year Computer Engineering student at National University of Singapore from August 2022. My research focuses on Efficient Generative AI.
🔥 News
-
2025.04: 🚀 TinyFusion is selected as CVPR 2025 Highlight.
-
2025.01: 🥳 Our team won the Second Place of 2025 SkiTB Visual Tracking Challenge.
-
2025.01: ⛵ Start my exchange journey at University of Washington!
-
2024.05: 🎉 PixelGen wins ACM/IEEE IPSN 2024 Best Demonstration Runner-Up award.
📝 Publications

TinyFusion: Diffusion Transformers Learned Shallow
CVPR’25 Highlight (3%) | Tiny DiTs at 7% Training Costs | 2x Faster Inference
Gongfan Fang*, Kunjun Li*, Xinyin Ma, Xinchao Wang (Equal-first author)
- Directly optimizes the recoverability of the pruned model, which ensures better performance after fine-tuning.
- Achieveing a 2x speedup using less than 7% of the original pre-training cost with a FID score of 2.86.

PixelGen: Rethinking Embedded Camera Systems for Mixed-Reality
| Novel embedded camera system for mixed reality
Kunjun Li, Manoj Gulati, Dhairya Shah, Steven Waskito, Shantanu Chakrabarty and Ambuj Varshney
- Generate High Resolution RGB images from Monochrome and sensor data.
- Novel representation of the surroundings from invisible signal.