Meitu Imaging and Vision Lab (MTlab) won first place in the Image Enhancement Challenge and third place in the Image Dehazing Challenge at the 2019 New Trends in Image Restoration (NTIRE) conference. The conference was held in conjunction with the 2019 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), the world's premier conference in the field of Artificial Intelligence (AI). Over 200 teams registered in both the Image Enhancement and Dehazing Challenges.
Using deep learning models based on visual aesthetics, the MTlab team was able to make the contest photo more visually appealing to the human eye with high fidelity. Contestants were ranked using fidelity and perceptual scores, as well as peoples' opinions on the quality of each processed image.
The Image Dehazing Challenge required teams to recover the content from an original image that was obscured by fog. By using various deep learning technologies such as Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN), MTlab team was able to effectively restore the original content of the image and place third overall in the content.
"The participating teams worked to solve the real-world photo enhancement problem of mapping low-quality photos from mobile phones to those captured by DSLR cameras," said MTlab Director Pengfei Wan. "There is a growing need to improve the quality of photos captured by lower-end mobile camera sensors: MTlab's extensive experience in perceptual image enhancement has made it a strong contender in this area."
The objectives of the NTIRE 2019 Challenge on Single Image Enhancement included gauging and pushing the state-of-the-art in image/video restoration and enhancement and comparing different approaches and solutions to different problems in the field. The challenge involved several computer vision subtasks, such as image denoising, image resolution, and sharpness enhancement, image color/contrast/exposure adjustment, and more. NTIRE is the largest and most significant event in the field of image/video restoration and enhancement: it attracts teams from top universities, research institutions, and corporations around the world.
MTlab is Meitu's core research and development department. Its areas of focus include computer vision (CV), deep learning (DL), and augmented reality (AR). MTlab continues to explore new possibilities in the field of image enhancement and has a proven track record of success in its field. Last September, MTlab placed first and significantly ahead of its competitors in terms of speed, subjective effect, and objective image quality at the PIRM Image Enhancement Challenge at the 2018 European Conference on Computer Vision (ECCV). This year, MTlab released the Meitu AI Open Platform (http://ai.meitu.com/index/) to provide individual developers and enterprises across multiple industries with market-proven professional AI algorithm services and solutions.
Moving forward, MTlab will continue to focus on its core AI technologies to deliver even more superior services for its users and corporate clients. In addition to image enhancement, MTlab is at the leading edge of facial detection, human keypoint detection, and image segmentation and generation. Meitu's self-developed AI algorithms have been widely applied to many of its product's features, including BeautyCam's AI Portrait, the Meitu APP's Anime Avatar, Meitu Smartphone's Super NightCam, and MEITUGENIUS' (also known as Meitu Magic Mirror) virtual makeup trial.