YijingYang

MCL Research Presented at ICASSP 2018

During 15-20 April, 2018, MCL member Junting Zhang presented her paper at 2018 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2018) in Calgary, Alberta, Canada. The title of the paper is “A Fully Convolutional Tri-branch Network (FCTN) for Domain Adaptation”, with Chen Liang and C.-C. Jay Kuo as co-authors. Here is an abstract of the paper:

“A domain adaptation method for urban scene segmentation is proposed in this work. We develop a fully convolutional tri-branch network, where two branches assign pseudo labels to images in the unlabeled target domain while the third branch is trained with supervision based on images in the pseudo-labeled target domain. The re-labeling and re-training processes alternate. With this design, the tri-branch network learns target-specific discriminative representations progressively and, as a result, the cross-domain capability of the segmenter improves. We evaluate the proposed network on large-scale domain adaptation experiments using both synthetic (GTA) and real (Cityscapes) images. It is shown that our solution achieves the state-of-the-art performance and it outperforms previous methods by a significant margin. ”

Congratulations to Junting for her successful presentation at ICASSP!

By |April 22nd, 2018|News|Comments Off on MCL Research Presented at ICASSP 2018|
Read More

MCL Research on Medical Image Segmentation

Biomedical image segmentation is a technique for automatic detection of organ boundaries within an image, used to obtain diagnostic insights in the field of medicine. Segmentation results are used in applications involving organ measurements, cell detection and blood flow simulations [1].

Zebrafish is a vertebrate that have similar organs and tissues as humans, making it a valuable model for studying human genetics and disease [2]. In this project, we use the zebrafish organs dataset for the analysis and evaluation of medical image segmentation methodologies.

Presently, we have trained a CNN based segmentation model using RefineNet [3] and a Saak [4] based model on the zebrafish data. Our goal is to compare and analyze the performance of CNN with Saak segmentation model. In future, we aim to improve these results by adopting content adaptive Saak with clustering techniques and statistical analysis algorithms.

-By Shilpashree Rao and Ruiyuan Lin

 

Reference:

[1]”Medical Image Segmentation”. [Online]. Available: https://www5.cs.fau.de/research/groups/medical-image-segmentation. [Accessed: 01- Apr- 2018].

[2]”Why use the zebrafish in research?”, 2018. [Online]. Available: https://www.yourgenome.org/facts/why-use-the-zebrafish-in-research. [Accessed: 01- Apr- 2018].

[3] G. Lin, A. Milan, C. Shen and I. Reid, “RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.

[4] C.-C. J. Kuo and Y. Chen, “On data-driven Saak transform,” Journal of Visual Communication and Image Representation, vol. 50, pp. 237–246, 2018.

Image credits:

Image showing the architecture of RefineNet is from [3].

Image showing the architecture of multi-stage Saak is from [4].

By |April 1st, 2018|News|Comments Off on MCL Research on Medical Image Segmentation|
Read More

MCL Research on Saak Transform

It is well known that CNNs-based methods have weaknesses in terms of efficiency, scalability, and robustness. CNNs-based methods require large computational efforts and are not scalable to the change of object class numbers and the dataset size. Furthermore, these CNN models are not robust to small perturbations due to their excess dependence on the end-to-end optimization methodology. The Saak (Subspace approximation with augmented kernels) transform [1] is proposed to provide the possible solution to overcome these shortcomings.

The Saak transform consists of two new ingredients on top of traditional CNNs. They are: subspace approximation and kernel augmentation. The Saak transform allows both forward and inverse transforms so that it can be used for image analysis as well as synthesis (or generation). One can derive a family of joint spatial-spectral representations between two extremes – the full spatial-domain representation and the full spectral-domain representation using multi-stage Saak transforms. Being different with CNNs, all transform kernels in multi-stage Saak transforms are computed by one-pass feedforward process. Neither data labels nor backpropagation is needed for kernel computation.

Currently, we have successfully developed Saak transform approaches [2] to solve the handwritten digits recognition problem. This new approach has several advantages such as higher efficiency than the lossless Saak transform, scalability against the variation of training data size and object class numbers and robustness against noisy images. In the near future, we would like to apply the Saak transform approach to the general object classification problem with more challenging datasets such as CIFAR-10, CIFAR-100, and ImageNet.

 

Reference

[1]  C-C Jay Kuo and Yueru Chen, “On Data-driven Saak Transform,” arXiv preprint arXiv:1710.04176, 2017.

[2] Chen, Y., Xu, Z., Cai, S., Lang, Y., & Kuo, C. C. J. (2017). A Saak Transform Approach to Efficient, Scalable [...]

By |February 25th, 2018|News|Comments Off on MCL Research on Saak Transform|
Read More

Welcome New MCL Member Zhaolei Xiao!

We are so happy to welcome a new Master’s member of MCL, Zhaolei Xiao. Here is an interview with Zhaolei.

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

My name is Zhaolei Xiao, come from the capital city of China, Beijing. I’m focusing on signal and image processing and this is my last semester in master’s program in USC. I like dealing with multimedia data: compress files, denoise a signal, and solve inpainting problem, etc., all of them give me a lot of fun. Moreover, after acquiring knowledge in machine learning field, like convolutional neural networks, building classifiers, I find it a new world to explore.

2. What is your impression about MCL and USC?

USC is the home of the Signal and Image Processing Institute (SIPI) and a leader over 40 years which is one of the reasons why I chose here to study. But not only that, professors are knowledgeable and helpful also in career, classmates are friendly and enthusiastic. I enjoy talking with them about academic and life.

I know MCL from Prof. Kuo who is my professor of the course Multimedia Data Compression that MCL likes a family, everyone helps each other and get progress together. I’m looking forward to making friends with all of them and doing cooperative research. It must be very pleasant.

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

Since I’m interested in compression and machine learning, Prof. Kuo’s project of joint compression and understanding catch my eyes. I treasure this opportunity very much and want to go deep into this application.

By |January 21st, 2018|News|Comments Off on Welcome New MCL Member Zhaolei Xiao!|
Read More