Dongmei Mo

My research interests are in computer vision, deep learning, optimization and machine learning with the goal of intelligent applications, intelligent fashion aesthetics cognition, personalized recommendation and creative generation.

Recent Highlights

[10 Dec. 2024] One of our papers has been accepted in AAAI 2025.

[Sep 2024] I'm going to be the guest lecturer in School of Fashion and Textiles at PolyU, to develop a new subject of Fashion Market Intelligence, using machine learning and AI methods for decision making in fashion businesses.

[Aug. 2024] I have been invited as a visiting researcher in the Institute of Industrial Automation and Software Engineering at University of Stuttgart, Germany.

Education

Sep. 2018-Feb. 2022 | Ph.D. | School of Fashion and Textile | The Hong Kong Polytechnic University

Sep. 2015-Jun. 2018 | M.A. | College of Computer Science and Software Engineering | Shenzhen University

Projects

Mar. 2022-Present, Postdoctoral Researcher, AI-based Fashion Design Assistant (AiDA) , the InnoHK Research Clusters.
Responsibility: mainly focus on developing aesthetic methods for collection generation and designer preference analysis. Meanwhile, collaborating with team members for prototype design and testing, and platform demonstration.
Achievement: the research outputs have now formed a design assistant for streamlining the process of developing fashion collections. The first collaboration of the assistant and designers was presented in a Fashion X AI fashion show in Dec. 2022 .

Oct. 2022-Present, Postdoctoral Researcher, Intelligent Fashion Aesthetic Evaluation System , the InnoHK Research Clusters.
Responsibility: mainly engaged in developing compatibility learning methods for personalized outfit recommendation and generation. Meanwhile, participating in the framework design of the prototype based on the research outputs.
Achievement: the research outputs have now formed a prototype and are ready for trial in the market for enhancing fashion retailing.

Teaching

Autumn 2020 Merchandising Management, The Hong Kong Polytechnic University

2019 Omni-Channel Fashion Marketing and Retailing, The Hong Kong Polytechnic University

Spring 2019 Management Principles in the Fashion Business, The Hong Kong Polytechnic University

Honors and Awards

[Mar. 2022] Research Talent Hub - Innovation and Technology Fund, Hong Kong

[Jun. 2018] Outstanding Postgraduate, Shenzhen University, China

[Oct. 2017] National Scholarship, China

  Research

My research covers image detection, classification and retrieval using traditional machine learning and deep learning algorithms.

Personalized Fashion Recommendation via Deep Personality Learning

This paper integrates user personality with physical attributes for fashion recommendation. The proposed personality learning model (P-Net) integrates user characteristics, including personality and personal physical information (such as skin colour, hair colour, etc.), with fashion styles for a personalized recommendation. P-Net first learns outfit embeddings via a feature encoder, and the embeddings are then fed to a message-passing network to model the relations among different outfits. The personality-style learning module learns the fashion personalities of the users, and the physical compatibility is learned by exploring the relations of the feature embeddings and the physical attributes via a Transformer module. Qualitative and quantitative results on a new stylish outfit of personality (SOP) dataset indicate the superiority of P-Net compared with state- of-the-art methods and discover the potential of the combination of fashion aesthetics and psychological science.

P-Net contains two components: fashion personality learning (the first pipeline) and physical compatibility learning (the second pipeline).

Private Stylists: Personalized Fashion Compatibility Learning

The work adopts physical and fashion attributes for effective personalized fashion compatibility evaluation and recommendation. The physical attributes are concluded into seven aspects: body shape, skin color, hairstyle, hair color, height, breast size (breasts), and color contrast. The model can not only predict the fashion attributes of the outfit’s top, bottom, shoes, and bag items, but also predict the incompatible physical attributes of an individual towards the given outfit. It can be used to recommend outfits that best fit an individual and the predicted fashion attributes can be used for result explanation.

The personalized compatibility evaluation framework. The framework composes of three parts: The image feature extractor, the first transformer encoder which is for fashion attribute prediction, and the second transformer encoder which is for physical attribute prediction.

Neural Stylist via Online Styling Service

Compatibility analysis from human aesthetics vs. automatic evaluation, diagnosis and recommendation. To proposes a visual and semantic representation model for explainable evaluation and recommendation.

The architecture of the proposed framework. The framework comprises four components: 1. multi-layer feature analysis; 2. colour- correlation enhancement; 3. visual-semantic similarity preserving, and 4: outfit diagnosis and recommendation.

Image Classification & Retrieval

The visualization of testing data on the Fashion-MNIST dataset. The image classification problem is basic when doing image interpretation research.

I study the problem of image retrieval and classification with Deep Hashing learning. Existing hashing methods are mainly focused on general images but few of them consider the application of fashion image retrieval and textile defect classification.

Image Detection

The results of locating multiple categories of defects for given fabric images. I design optimization models for the specific problems and find the optimal solution for the designed objective functions. The experiments are evaluated with the proposed iterative algorithms.

Use grid search to find optimal values for three parameters.

Low-rank decomposition problem solved by algorithm based on Alternating Direction Method of Multipliers (ADMM).

Feature Extraction & Dimensionality Reduction

I studied feature extraction and dimensionality reduction during my postgraduate period and published several papers on IEEE Transactions Q1 journals.

Presentation of reconstruction images and images of the first 2 projections from subspace.

Verify the performance on keeping the local structure of the data. These are the results that the images mapped onto the two-dimensional plane described by two coordinates of the proposed method.

Manifold Learning

I develop algorithms to maintain the local geometric structure of the image samples (i.e. manifold structure of the images).

The result of image samples from two classes projected onto the subspace.

The algorithms can guarantee that the data projected on the learned subspace is of great discrimination. Thus, any samples projected onto the subspace can obtain maximal discrimination and thus the intrinsic local geometric structure of the data can be preserved.

   Publications

Working papers

[1] D. Mo, D. Souza, X. Zou, W. Wong, M. Deisenroth. "Fashion brand analysis: from classification to identity fuzziness learning."
[2] D. Mo, D. Souza, X. Zou, W. Wong, M. Deisenroth. "Exploring Gaussian Process Models for Advanced Textual-Visual Understanding: An Empirical Study on Fashion Design."
[3] D. Mo D. Souza, X. Zou, F. Pfaff, W. Wong, M. Deisenroth. "AI in Branding Fashion Design: Learning Historical Characteristics for Innovative Generation."

Selected Publications

[1] L. Shuai, D. Mo and W. Wong. "REB: Reducing biases in representation for industrial anomaly detection." Knowledge-Based Systems 290 (2024): 111563. [Code]

[2] D. Mo, X. Zou, and W. Wong. "Super stylist: personalized fashion recommendation via deep personality learning", The British Machine Vision Conference (BMVC) 2023. [Dataset]

[3] D. Mo, Z. Lai, J. Zhou, et al. "Scatter matrix decomposition for jointly sparse learning[J]". Pattern Recognition, 2023, 140: 109485.

[4] D. Mo, X. Zou, and W. Wong. "Towards private stylists via personalized compatibility learning", Expert Systems with Applications (2023): 119632.

[5] D. Mo, X. Zou, and W. Wong. "Neural stylist: Towards online styling service[J]", Expert Systems with Applications, 2022: 117333.

[6] D. Mo, W. Wong, Z. Lai. "Weighted double low-rank decomposition with application to fabric defect detection[J]", IEEE Transactions on Automation Science and Engineering 2020, (18.3): 1170-1190.

[7] D. Mo, X. Liu, Y. Ge and W. Wong. "Concentrated hashing with neighborhood embedding for image retrieval and classification[J]", International Journal of Machine Learning and Cybernetics, 2022, 13(6): 1571-1587.

[8] D. Mo, Z. Lai, W. Wong. "Jointly sparse locality regression for feature extraction[J]", IEEE Transactions on Multimedia, 2019, 22(11): 2873-2888.

[9] D. Mo, and Z. Lai. "Robust jointly sparse regression with generalized orthogonal learning for image feature selection[J]", Pattern Recognition, 2019, 93: 164-178.

[10] D. Mo, Z. Lai, and W. Wong. "Locally joint sparse marginal embedding for feature extraction[J]", IEEE Transactions on Multimedia, 2019, 21(12): 3038-3052. [Code]

[11] Z. Lai, D. Mo, et al. "Robust discriminant regression for feature extraction[J]", IEEE Transactions on Cybernetics, 2017, 48(8): 2472-2484.

[12] Z. Lai, D. Mo, Wen J, et al. "Generalized robust regression for jointly sparse subspace learning[J]". IEEE Transactions on Circuits and Systems for Video Technology, 2018, 29(3): 756-772.