Research Assistant Professor
Laboratory: Multimedia Lab

Room 717
Ho Sin-Hang Engineering Building
Department of Information Engineering
The Chinese University of Hong Kong
Shatin, NT, Hong Kong

ccloy at
ccloy at

Chen Change Loy is a Research Assistant Professor in the Chinese University of Hong Kong. He received his PhD (2010) in Computer Science from the Queen Mary University of London (Vision Group). From Dec. 2010 – Mar. 2013, he was a postdoctoral researcher at Queen Mary University of London and Vision Semantics Limited. His research interests include computer vision and pattern recognition, with focus on face analysis, deep learning, and visual surveillance. He serves as an Associate Editor of IET Computer Vision Journal. He also serves as regular reviewer for top-ranking journals (TPAMI, IJCV, TIP, TCSVT, PR) and conferences (CVPR, ECCV, ICCV, ACCV, BMVC).

Google scholar profile





Chen Change Loy
PolyNet: A novel design of ultra-deep networks by CU-DeepLink. Project page.
Our PolyNet design yields higher accuracy than Inception-ResNet given the same computation budget.
Computer Vision and Image Understanding Special Issue on Image and Video Understanding in Big Data

NIPS 2016

Local Similarity-Aware Deep Feature Embedding (NIPS 2016)
PDF Technical Report

We introduce Position-Dependent Deep Metric (PDDM) unit pluggable to any convolutional networks for automated hard samples mining leading to more effective deep feature embedding learning.

ECCV 2016

Accelerating the Super-Resolution Convolutional Neural Network (ECCV 2016)
PDF Technical Report Supplementary Material Project Page | Codes

We redesign SRCNN to achieve a speed up of more than 40 times with even superior restoration quality. The new FSRCNN can run in real-time on a generic CPU. Read more ...

Joint Face Representation Adaptation and Clustering in Videos (ECCV 2016)
PDF Project Page

Deep learning with adaptation for face clustering in video. State-of-the-art results are reported on Accio (Harry Potter), BF0502, and Notting-Hill datasets. Read more ...

Depth Map Super Resolution by Deep Multi-Scale Guidance (ECCV 2016)
PDF Project Page

A new multi-scale guidance architecture for depth-map super-resolution. Top results on Middlebury RGBD datasets. Read more ...

Deep Specialized Network for Illuminant Estimation (ECCV 2016)
PDF Project Page

State-of-the-art results on Color Checker and NUS 8-camera datasets. Using a novel deep architecture that consists of two networks - one to output multiple hypotheses, and another one for hypothesis selection. A diversity-encouraging "winner-take-all" learning scheme is proposed to train the specialized network. Read more ...

Deep Cascaded Bi-Network for Face Hallucination (ECCV 2016)
PDF Technical Report Project Page | Codes

A novel framework for hallucinating faces of unconstrained poses and with very low resolution (face size as small as 5pxIOD). Read more ...

Human Attribute Recognition by Deep Hierarchical Contexts (ECCV 2016)
PDF Project Page

We present a new approach that exploits deep hierarchical contexts for accurate human attribute recognition. The model achieves top results on Berkeley Attributes of People and HAT datasets. A new large-scale WIDER Attribute dataset with 14 human attribute labels and 30 event class labels is introduced. Read more ...

CVPR 2016

WIDER FACE: A Face Detection Benchmark (CVPR 2016, Oral)
PDF Technical Report Leaderboard | Dataset

We show that there is a gap between current face detection performance and the real world requirements. To facilitate future face detection research, we introduce the WIDER FACE dataset, which is 10 times larger than existing datasets. The dataset contains rich annotations, including occlusions, poses, event categories, and face bounding boxes. Read more ...

Unconstrained Face Alignment via Cascaded Compositional Learning (CVPR 2016)
PDF Project Page

To equip cascaded regressors with the capability to handle global shape variation and irregular appearance-shape relation in the unconstrained scenario, we partition the optimisation space into multiple domains of homogeneous descent, and predict a shape as a composition of estimations from multiple domain-specific regressors. Read more ...

Learning Deep Representation for Imbalanced Classification (CVPR 2016, Spotlight)
PDF Project Page

We demonstrate that more discriminative deep representation can be learned by enforcing a deep network to maintain both inter-cluster and inter-class margins. This tighter constraint effectively reduces the class imbalance inherent in the local data neighborhood. We show that the margins can be easily deployed in standard deep learning framework through quintuplet instance sampling and the associated triple-header hinge loss. Read more ...

Unsupervised Learning of Discriminative Attributes and Visual Representations (CVPR 2016)
PDF Project Page

While most attribute learning methods are supervised by costly human-generated labels, we introduce a simple yet powerful unsupervised approach to learn and predict visual attributes directly from data. Given a large unlabeled image collection as input, we train deep Convolutional Neural Networks (CNNs) to output a set of discriminative, binary attributes often with semantic meanings. The visual representations learned in this way are also transferrable to other tasks such as object detection. Read more ...

Slicing Convolutional Neural Network for Crowd Video Understanding (CVPR 2016, Spotlight)
PDF Project Page

We propose a novel spatio-temporal CNN, named Slicing CNN (S-CNN), based on the decomposition of 3D feature maps into 2D spatio- and 2D temporal-slices representations. Read more ...

Sketch Me That Shoe (CVPR 2016, Oral)
PDF Project Page

We investigate the problem of fine-grained sketch-based image retrieval (SBIR), where free-hand human sketches are used as queries to perform instance-level retrieval of images. We introduce a new database of 1,432 sketch-photo pairs from two categories with 32,000 fine-grained triplet ranking annotations. We then develop a deep triplet-ranking model for instance-level SBIR with a novel data augmentation and staged pre-training strategy to alleviate the issue of insufficient fine-grained training data. Read more ...