News

We are so happy to welcome a new Post Doctor, Dr Xinfeng Zhang, this Fall 2017. Let us hear what he has to say about his research and MCL.

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

My name is Xinfeng Zhang, and I received my B.S. degree in computer science from Hebei University of Technology, Tianjin, China, in 2007, and the Ph.D. degree in computer science from the Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China, in 2014. From, Jul. 2014 to Oct. 2017, I was a research fellow in Rapid-Rich Object Search (ROSE) Lab in Nanyang Technological University, Singapore. My research interests include image/video compression, processing and quality assessment. I am also interested in image/video retrieval and analysis.

2. What is your impression about MCL and USC?

When I joined MCL, I was impressed with the professional group seminar, which is very formal and beneficial for both the speakers and audience. The active discussions among students also provide various views for us to think about questions. Moreover, I love the environment of USC, and it is real very beautiful.

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

I am very interested in the Saak transform, and I think it is a new powerful tool in image/video compression and understanding. Therefore, I hope to explore the characteristics of Saak transform, and investigate the higher compression performance using Saak transform. I also hope to become friends with all the other MCL members.

MCL Student Junting Zhang presented a paper at the 5th IEEE Global Conference on Signal and Information Processing (GlobalSIP 2017) in Montreal Quebec, Canada on November 15, 2017. Here’s an abstract of the paper :
Scene text detection is a critical prerequisite for many fascinating applications for vision-based intelligent robots. Existing methods detect texts either using the local information only or casting it as a semantic segmentation problem. They tend to produce a large number of false alarms or cannot separate individual words accurately. In this work, we present an elegant segmentation-aided text detection solution that predicts the word-level bounding boxes using an end-to-end trainable deep convolutional neural network. It exploits the holistic view of a segmentation network in generating the text attention map (TAM) and uses the TAM to refine the convolutional features for the MultiBox detector through a multiplicative gating process. We conduct experiments on the large-scale and challenging COCO-Text dataset and demonstrate that the proposed method outperforms state-of-the-art methods significantly.

Let us hear what he has to say about his defense and an abstract of his thesis.

Online object tracking is one of the fundamental computer vision problems. It is commonly used in real world applications such as traffic control and safety in video surveillance, autonomous vehicle, robotic navigation, medical imaging, etc. It is a very challenging problem due to multiple time-varying attributes in video sequences.

In this research, we investigate two different kinds of tracking problems: single object tracking (SOT) and multiple object tracking (MOT). First, we attempt to achieve online single object tracking using both spatial and motion cues with two novel methods. One is a traditional framework called “temporal prediction and spatial refinement (TPSR)” tracker, consisting of three cascaded modules: pre-processing (PP), temporal prediction (TP) and spatial refinement (SR). Another one is based on convolutional neural network architecture that treats the tracking as a detection problem, called ”Motion-Guided Convolutional Neural Network (MGNet) Tracker”. It has two innovations: 1) adoption of a motion-guided candidate selection (MCS) scheme based on a dynamic prediction model, and 2) usage of a RGB- plus-motion 5-channel input to the convolutional neural network (CNN). From the two proposed methods, we have showed the advantages of combining spatial information and motion cue together to improve the tracking performance.

Second, from the proposed SOT technique, we build an online multiple object tracking system with advanced model update and matching. This method treats the MOT problem as an online tracking problem, rather than the global optimization framework. There are three major components in this tracking system: 1) a system platform built upon multiple CNN single object trackers in MOT environment; 2) the proposed advanced online update strategy including incremental and aggressive update mode; 3) [...]

Congratulations to Eddy Wu for passing his defense. His thesis title is “Deep Learning Techniques for Supervised Pedestrian Detection and Critically-Supervised Object Detection” with three major topics, as follow:

In the first topic, a boosted convolutional neural network (BCNN) system is proposed to enhance the pedestrian detection performance. Being inspired by the classic boosting idea, we develop a weighted loss function that emphasizes challenging samples in training a convolutional neural network (CNN). Two types of samples are considered challenging: 1) samples with detection scores falling in the decision boundary, and 2) temporally associated samples with inconsistent scores. Finally, we train a boosted fusion layer to benefit from the integration of these two weighting schemes. We test the corresponding BCNN on the Caltech pedestrian dataset in the experiment and observe a significant performance gain over the Fast-RCNN baseline.

In the second topic, a semi-supervised learning method is proposed for pedestrian detection in a domain adaptation setup. The proposed clustered deep representation adaptation (CDRA) method uses a small amount of labeled data to train an intial detector, extracts the deep representation and, then, clusters samples based on the space spanned by the deep representation. A purity measurement mechanism is applied to each cluster to provide a confident score to the estimated class of unlabeled data. Along with a weighted training approach, the CDRA method is shown to achieve the state-of-the-art performance against some large scale datasets.

In the third topic, we propose a new framework called critically-supervised learning that mimics children learing behaviors. Several novel components are proposed to fulfill the high level concept, including negative object proposal, critical example mining, and a machine-guided labeling process based on question answering. A labeling time model is proposed to [...]