Selected Publications

This paper presents a simple yet effective approach that for the first time enables transferring arbitrary styles in real-time. At the heart of our method is a novel adaptive instance normalization (AdaIN) layer that aligns the mean and variance of the content features with those of the style features. Our method achieves speed comparable to the fastest existing approach, without the restriction to a pre-defined set of styles. In addition, our approach allows flexible user controls such as content-style trade-off, style interpolation, color & spatial controls, all using a single feed-forward neural network.
ICCV 2017

This paper proposes a novel generative model named Stacked Generative Adversarial Networks (SGAN). The core idea is to leverage the powerful discriminative representations to guide a generative model. Our model consists of a top-down stack of GANs, each learned to generate lower-level representations conditioned on higher-level representations. We also introduce a conditional loss and a novel entropy loss that improve quality and diversity respectively. Based on visual inspection, Inception scores and visual Turing test, we demonstrate that SGAN is able to generate images of higher quality than GANs without stacking.
CVPR 2017

This paper proposes a novel framework for structured labeling problems. We develop a convolutional pseudo-prior (ConvPP) network which can capture the contextual dependency between ground-truth labels. A novel fixed-point training strategy is employed to facilitate an end-to-end learning process. Experiments on sequential labeling and image labeling benchmarks demonstrate the superior performance of the proposed method.
ECCV 2016

In this paper, we present a focused study to narrow the semantic gap in saliency prediction with an architecture based on Fully Convolutional Networks. Through extensive experiments, we reveal important architectural choices that lead to substantial improvements. We compare our method with 14 saliency models on 6 public eye tracking benchmarks and show a performance surpassing by a big margin the state-of-the-art.
ICCV 2015
  • Full list of publications available at Google Scholar.
  • Teaching

    Teaching Assistant:


    Here are some interesting facts about myself:

    • I am an amateur Go player and a member of the Cornell Go Club. In 2016-2017 academic year, I am participating in ACGL (American Collegiate Go League), representing Cornell University. I am (officially) a 2-dan amateur Go player and a 4-dan player on Eweiqi.
    • I love reading books, especially science fictions and philosophy books. I am particularly interested in the contemporary Analytic Philosophy and more specifically the Philosophy of Mind, though I can hardly understand philosophies before Age of Enlightenment.
    • I am a music lover. My favorite genres include classical music, new-age music, and Cantonese popular songs. I can play the piano at the beginner level and am still learning it. I love classical pieces composed by Schubert, Bach, and Chopin.
    • I am an INTJ, according to MBTI assessment.