Publications

	StableKD: Breaking Inter-block Optimization Entanglement for Stable Knowledge Distillation Shiu-hong Kao*, Jierun Chen*, S.-H. Gary Chan Under review. *Equal contribution. [arXiv] [Code] We reveal the issue of Inter-block Optimization Entanglement (IBOE) in end-to-end KD training and further propose StableKD to stablilize optimization. Extensive experiments show StableKD achieve high accuracy, fast convergence, and high data efficiency.
	InceptionHuman: Controllable Prompt-to-NeRF for Photorealistic 3D Human Generation Shiu-hong Kao, Xinhang Liu, Yu-Wing Tai, Chi-Keung Tang Under review. Code coming soon. [arXiv] We propose InceptionHuman, a NeRF-based generative framework incorporating state-of-the-art diffusion models, which receives any types and any sizes of prompts, (e.g. text, pose, style) to generate realistic 3D human.
	Diffusion-Generated Pseudo-Observations for High-Quality Sparse-View Reconstruction Xinhang Liu, Jiaben Chen, Shiu-hong Kao, Yu-Wing Tai, Chi-Keung Tang European Conference on Computer Vision (ECCV), 2024 [Project page] [arXiv] We introduce Deceptive-NeRF/3DGS, a new method for enhancing the quality of reconstructed NeRF/3DGS models using synthetically generated pseudo-observations, capable of handling sparse input and removing floater artifacts.
	Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks Jierun Chen, Shiu-hong Kao, Hao He, Weipeng Zhuo, Song Wen, Chul-Ho Lee, S.-H. Gary Chan IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023 [Paper] [Code] We propose a simple yet fast and effective partial convolution (PConv), as well as a latency-efficient family of network architectures called FasterNet.
	Integer Partition: Further Discussion of the Number of the Integer-side Triangles Shiu-hong Kao, Pohsing Chou, Gerard Jennhwa Chang Yau's Award, National Taiwan University, 2018 [中文論文] [English paper] We prove the number of integer-side triangles/quadrilaterals with fixed perimeters in two different ways and deduce a recursive relaitonship for n-sided polygons.

Projects

• Dynamic Neural Network Comparison: Efficiency and Performance
  Shiu-hong Kao, Jierun Chen, S.-H. Gary Chan
  Undergraduate Research Opportunities Program, Fall 2021, HKUST
  [pdf]
  

  We test, analyze, and compare the performance and efficiency across different deep learning models.



• Press Release Classification on Mobile Application
  Shiu-hong Kao, Dimitris Chatzopoulos
  Undergraduate Research Opportunities Program, Spring 2021, HKUST
  [pdf]
  

  We develop a text-classifying AI model compact with memory-limited mobile devices.



• AlignKD: A Low-cost Technique for Convolutional Shortcut Removal
  Shiu-hong Kao, Bo-rong Lai
  Course project for COMP4471 Deep Learning in Computer Vision, HKUST
  [pdf]
  

  AlignKD is a cheap method to remove shortcuts in convolutional neural networks while preserving the performance.