News

The objective of image-to-image (I2I) translation involves learning a mapping from a source domain to
a target domain. Specifically, it aims at transforming images of the source style to those of the target
style with content consistency. While there is a domain gap, it can be mitigated by aligning the
distributions of the source and the target domains. Nevertheless, disparities between class distributions
of the source and target domains result in semantic distortion (see Figure 1); namely, different
semantics of correspondent regions between input and output. The semantic distortion could potentially
impact the efficacy of downstream tasks, such as semantic segmentation or object classification.
In this work, we propose a novel contrastive learning-based method that alleviates semantic
distortion by ensuring semantic consistency between input and output images. This is achieved by
enhancing the inter-dependence of structure and texture features between input and output by
maximizing their mutual information. In addition, we exploit multiscale predictions to boost the
I2I translation performance by employing global context and local detail information jointly to
predict translated images of superior quality, especially for high-resolution images. Hard negative
sampling is also applied to reduce semantic distortion by sampling informative negative samples.
For brevity, we refer to our method as SemST. Experiments conducted on I2I translation across
various datasets demonstrate the state-of-the-art performance of the SemST method. Additionally,
utilizing refined synthetic images in different UDA tasks confirms its potential for enhancing the
performance of UDA.

Congratulations to Zhiruo Zhou for passing her defense on September 25, 2023. Zhiruo’s thesis
is titled “Green Unsupervised Single Object Tracking: Technologies and Performance
Evaluation.” Her Dissertation Committee includes Jay Kuo (Chair), Antonio Ortega, and Stefanos
Nikolaidis (Outside Member). Zhiruo received several questions and suggestions from the
Committee members. Zhiruo answered the questions professionally.
Congratulations to Zhiruo for this milestone moment in life. MCL News team invited Zhiruo for a
short talk on her thesis and PhD experience, and here is the summary. We thank Zhiruo for her
kind sharing, and wish her all the best in the next journey.
“Video object tracking is one of the fundamental problems in computer vision and has diverse
real-world applications such as video surveillance and robotic vision. Given the ground-truth
bounding box of the object in the first frame of a test video, a single object tracker (SOT)
predicts the object box in all subsequent frames. Supervised trackers trained on labeled data
dominate the single object tracking field for superior tracking accuracy. The labeling cost and
the huge computational complexity hinder their applications on edge devices. Unsupervised
learning methods have also been investigated to reduce the labeling cost, but their complexity
remains high.
In my dissertation, I investigate the feasibility of lightweight high-performance tracking with
algorithmic transparency and no offline pre-training. I present our design of the green object
tracker that exploits spatial and temporal correlations at different granularities for more robust
tracking. It has been examined on a variety of benchmarks against recent state-of-the-arts, with
the inference complexity that is between 0.1%-10% of neural-network-based trackers. The
tracker is called ‘green’ due to its low computational complexity in both training and inference
stages, leading to a low [...]

We are so happy to welcome a new visiting Student, Teru Nagamori, joining MCL this fall. Here is a quick interview with Teru:

1. Could you briefly introduce yourself and your research interests?

I’m Teru Nagamori from Japan. I’m a master’s student at Tokyo Metropolitan University in Tokyo, Japan, and I’m visiting MCL for a month as a short-term exchange student. My research interest is to protect machine learning (ML) models from exploitation and to preserve personal visual information (e.g. human’s face, car license plate) on images used for training and testing ML models. These fields are called access control and privacy-preserving.
Also, I’ve worked at a company as a software engineer intern for a year and a few months, so I like to develop web services with Python, Vue, React, and so on.

2. What is your impression about MCL and USC?

First of all, I felt USC’s campus is so huge and beautiful. I wanted to spend the rest of my master’s program on this campus. I also felt MCL is a good environment too. Members are so kind, and each research is so interesting. While I’m staying here, I would like to take a bunch of knowledge and use it for my research when I return to Japan.

3. What is your future expectation and plan in MCL?

I have just a month to stay here, so what I can do is limited. So, I’ll research face recognition with encrypted images to expand my research. In addition, I would like to know many things from other members about not only things related to research but also culture and so on through having lunch or hanging out with them.

We are so happy to welcome a new graduate member, Catherine Aurelia Christie Alexander, joining MCL this summer. Here is a quick interview with  Catherine:

1. Could you briefly introduce yourself and your research interests?

My name is Catherine Aurelia Christie Alexander. I am a graduate student at USC, pursuing my Master’s in Electrical Engineering, specializing in Signal and Image Processing. My research interests include Machine Learning, Computer Vision, Signal Processing and Biomedical Imaging. 

2. What is your impression about MCL and USC?

 USC’s diversity and huge network is very impressive. There are many resources on and off campus that are very helpful and help one grow into an individual ready to face the world. MCL is one such place that has seen the growth of many students. The lab is full of enthusiastic, focused, and goal-driven students guided by Professor Kuo, who is very dedicated and inspiring. There is so much to learn from each person. Seminar sharing sessions give an overview of research in the lab and are very informative and encouraging. 

3.What is your future expectation and plan in MCL?

As a continuation of my summer research, I will be working with biomedical images, specifically on nuclei segmentation, and hope to develop skills and knowledge in this field. I would also like to contribute green AI solutions to the field of healthcare as a part of MCL. I look forward to connecting with all the members and having a great learning experience!

We are so happy to welcome a new graduate member,  Zijing Chen, joining MCL this summer. Here is a quick interview with Zijing:

1. Could you briefly introduce yourself and your research interests?

My name is Zijing Chen, and I am a second year graduate student here at USC pursuing an Electrical Engineering MS degree, with a focus on Machine Learning and Data Science. In the past, I have created and participated in several projects involving machine learning, such as traffic anomaly analysis and data fraud detection projects. Apart from academics, I enjoy playing music and painting during my free time.

2. What is your impression about MCL and USC?

MCL is an impressive lab consisting of many intelligent and approachable people, fostering a friendly and supportive environment. The amount of help I have received from the lab members and the guidance from Professor Kuo allows for a good research and learning experience. I look forward to knowing more about MCL members and working with them in the future.

3. What is your future expectation and plan in MCL?

So far, and for the rest of this summer, I will be researching demosaicing using green learning and the development of the LNT3 model. I look forward to making connections with other MCL members and facing more challenges in the future.