Deep learning is one of the most fascinating learning techniques. MMLAB
novel deep models for various vision tasks such as face recognition, face
pedestrian detection, and person parsing.
Automatically understanding the behaviors of crowd from video sequences is of
interest to the computer vision community, and has drawn more and more
in recent years. It has important applications to event recognition, traffic
flow estimation, behavior prediction, abnormality detection, and crowd
Face Sketch Synthesis and Recognition
Artists have a fascinating ability to capture the most distinctive
of human faces and depict them on sketches. Although sketches are very
from photos in style and appearance, human beings often can easily recognize
a person from his sketch. We design computer programs to also have such
Photo Quality Assessment based on Aesthetic Perception
MMLAB develops fast and effective computing algorithms which can
aesthetic assessment of large scale photo datasets inspired by the criteria
skills adopted by professional photographers.
One-Click Internet Image Search
We target on greatly improving user experience of web-based image search
with extremely simple user interaction. It interprets users’ search
with the minimum feedback effort: only one click on the query image from a
retrieved by text-based search.
Face Detection, Alignment, and Recognition
MMLAB develops novel algorithms for automatically locating facial keypoints,
face images, and identifying or verifying a person from a digital image or a
The CUHK Multimedia Lab (MMLab) is one of the pioneering institutes on deep learning. In GPU
Technology Conference (GTC)
2016, a world-wide technology summit, our lab is recognized as one of the top ten AI
and listed together with top research groups in the world (e.g. MIT, Stanford, Berkeley, and
Univ. of Toronto). Today, we remain one of the most active research labs in computer vision
deep learning, publishing over 40 papers on top conferences (CVPR/ICCV/ECCV/NIPS) every
Our lab has a large group of talented students, plenty of computational resources, and
financial support, and free research environment.
- The Multimedia Laboratory of the Department of Information Engineering is established by
Xiaoou Tang in July 2001.
- We won the CVPR 2009 Best Paper Award. This is the first one ever from Asia.
K. He, J. Sun, and X. Tang, "Single Image Haze Removal Using Dark Channel Prior
- Best paper awards by our lab's alumni:
- Dahua Lin with his paper "Construction of Dependent Dirichlet Processes based on
Processes", NIPS, 2010
- Dong Xu with his paper "Visual Event Recognition in Videos by Learning from Web
- Huan Wang with his paper "Exact Recovery of Sparsely-Used Dictionaries", COLT,
- Shuicheng Yan with his papers "Dynamic Captioning: Video Accessibility
Hearing Impairment" in ACM MM, 2010; "Automated Assembly of Shredded Pieces from
Multiple Photos" in ICME, 2010; "Wow! You Are So Beautiful Today!", ACM MM,
"Attributes-augmented Semantic Hierarchy for Image Retrieval", ACM MM, 2013
- Lab members that pursue oversea studies: MIT (4 students), Standford (2 students),
(1 student), Columbia (2 students), CMU (2 students).
- 15 Hong Kong PhD fellows.
- Our work with
Microsoft Research Asia (MSRA) are now being used by
two major Microsoft products.
- Intentsearch: being used in "Bing" for similar image search.
- Digital Effect: being used in MS webcam.
We released and updated two open source toolboxes MMDetection (object detection and
instance segmentation) and MMAction
(action recognition and detection).
WIDER Face and Person Challenge
We organize the second WIDER Face and Person
Challenge in conjunction with ICCV 2019. There
are four exciting tracks with great prizes. Deadline of challenge: July 25, 2019.
NTIRE 2019 winner
Our new video restoration method, EDVR, won all four tracks in the
NTIRE 2019 video
restoration and enhancement challenges.
We have 33 papers (7 oral) accepted to CVPR 2019.
Our lab have 3 papers accepted to ICLR 2019.
Our lab have 6 papers (4 oral and 1 spotlight) accepted to AAAI 2019.
COCO 2018 Challenge Winner
Our team MMDet wins the COCO
2018 Challenge (object detection/instance segmentation track). Codebase is released.
Our lab have 5 papers accepted to NeurIPS 2018.
Our lab have 26 papers accepted to ECCV 2018.
Our lab have 28 papers (1 oral and 7 spotlight) accepted to CVPR 2018.