Our Research

Research

MMLab@NTU

Members in MMLab@NTU conduct research primarily in low-level vision, image and video understanding, creative content creation, 3D scene understanding and reconstruction.

Super-Resolution

Our team is the first to introduce the use of deep neural networks to directly predict super-resolved images. Our journal paper on image super-resolution was selected as the `Most Popular Article' by IEEE Transactions on Pattern Analysis and Machine Intelligence in 2016. It remains as one of the top articles to date. Popular image and video super-resolution methods developed by our team include SRCNN, ESRGAN, EDVR, GLEAN, BasicVSR and CodeFormer.

Content Editing and Creation

We research new methods for generating high-resolution, realistic and novel contents in images and videos. We are also interested in investigating fundamental concepts in generative models. Some of our works include diffusion models, face generation and editing, high-resolution portrait video style transfer, video synthesis, human image generation, and style transfer.

Image and Video Understanding

We explore effective and efficient methods to detect, segment and recognize objects in complex scenes. We were the champion in COCO 2019 Object Detection Challenge 2019, and Open Images Challenge 2019. Some of our recent projects focus on open-vocabulary detection and segmentation, learning universal visual representation, multitasks models, and building multimodality models such as Otter.

3D Generative AI

Our team has been working on various tasks related to 3D generative AI to enable the creation of high-quality 3D models. Check out our recent work on novel view syntensis with free camera trajectories, OmniObject3D - a large vocabulary 3D object dataset with massive high-quality real-scanned 3D objects, algorithms to generate immersive 3D scenes, 3D avatar generation, and high-quality shape and texture reconstruction.

Deep Learning

We investigate new deep learning methods that are more efficient, robust, accurate, scalable, transferable, and explainable. We have been working on various problems like domain generalization, knowledge distillation, long-tailed recognition, and self-supervised learning. We participated in Facebook AI Self-Supervision Challenge 2019, and we won champions of all four tracks. Check out MMSelfSup, our open-source project for self-supervised learning.

Media Forensics

The popularization of Deepfakes on the internet has set off alarm bells among the general public and authorities, in view of the conceivable perilous implications. We have proposed two large-scale datasets for face forgery detection. See DeeperForensics and ForgeryNet. Also check out our recent work on detecting multimodal media manipulation.

Research

Video Demos

codeformer
CodeFormer (Hugging Face Demo)
vtoonify
VToonify (Hugging Face Demo)
scenedreamer
Otter Chatbot (Demo)
stylegan_human
Human Image Generation (Hugging Face Demo)
scenedreamer
SceneDreamer (Hugging Face Demo)
humman
4D Human Sensing
text2human
Text2Human (Hugging Face Demo)
bailando
Dance Generation
pastiche_master
Portrait Style Transfer (Hugging Face Demo)
esrgan
Anime Super-Resolution (Credit: Jan Bing)
final_fantasy
Game Super-Resolution (Credit: Snouz)
talkedit
Face Editing
dni
Effect Interpolation
animeinterp
Anime Video Interpolation
inpainting
Video Inpainting
zerodce
Low-Light Enhancement (Hugging Face Demo)
gan_prior
Image Restoration and Manipulation
liteflownet
Optical Flow Estimation
talking_head
Audio-Driven Face Reenactment

Selected

Media Coverage | Blog

Source: @Luke_Aaron