Xingxing Zou’s research while affiliated with The Hong Kong Polytechnic University and other places

... To address these limitations, researchers have explored various inversion techniques for GANs and DMs, including GAN inversion [25,33,40,48,51,54,55] and DDIM/DPM inversion [20,43], which estimates the initial random noise in DMs by reversing the U-Net denoising process [14,38]. Null-Text Inversion, for instance, guides the diffusion model toward consistent edits by performing a pivotal inversion combined with a null-text optimization, improving edit fidelity [30]. ...

Reference:

Training-Free Consistency Pipeline for Fashion Repose

... Subsequently, researchers [29], [30], [31], [32] utilized collaborative filtering to model user preferences based on user interaction data for personalized recommendation problem. The third stage incorporates contextual information to better understand users' needs, where researchers consider either user characteristics (e.g., physical attributes [33], [34], hair-style [35], [36]) or environmental factors (e.g., occasion [37], weather [38]) to deliver more targeted recommendations. In practical applications, existing outfit recommendation systems [39], [30], [40], [7] typically operate as ranking-based methods. ...

Reference:

FashionM3: Multimodal, Multitask, and Multiround Fashion Assistant based on Unified Vision-Language Model

... The creation of realistic virtual human images constitutes a pivotal area of research within the domains of computer vision and generative models. Among these, Pose-Guided Person Image Synthesis (PGPIS) is particularly focused on generating images that preserve the visual characteristics of a source image while adapting it to a specified target pose [14]. This technique finds applications in various fields, including virtual reality (VR) and ecommerce [28], and holds significance for tasks such as person re-identification [35] and the generation of sign language videos. ...

Reference:

Real-Time Person Image Synthesis Using a Flow Matching Model

... Simulation-based datasets [Bertiche et al. 2020;Black et al. 2023;Gundogdu et al. 2019;Jiang et al. 2020,?;Narain et al. 2012;Patel et al. 2020;Santesteban et al. 2019;Xiang et al. 2020;Zou et al. 2023] often include simulation-based datasets [Bertiche et al. 2020;Black et al. 2023;Gundogdu et al. 2019;Jiang et al. 2020;Narain et al. 2012;Patel et al. 2020;Santesteban et al. 2019;Xiang et al. 2020;Zou et al. 2023], which use physics engines to simulate and enhance the physical plausibility of synthetic 3D garments. While these datasets are more efficient to produce than 3D scanning datasets, they generally suffer from limited garment style diversity, poor garment deformation, and low-quality paired images, reducing their practical use for realworld image data tasks. ...

Reference:

GarmageNet: A Dataset and Scalable Representation for Generic Garment Modeling

... For faster recognition, Kaur et al. [23] proposed a framework that combines CNNs with SURF for efficient clothing recognition. Zhu et al. [24] proposed sRA-Net to accurately obtain attributes representations by utilizing multiple latent relationships in clothing images to improve the performance of fashion attributes recognition. This paper treats fine-grained clothing attributes recognition as an object detection task. ...

Reference:

A fine-grained attributes recognition model for clothing based on improved the CSPDarknet and PAFPN network

... In contrast, some researchers have aimed to capture a holistic view of outfit-level representations through bidirectional LSTMs ( [5]) or graph neural networks ( [4], [10]). Some studies also have shifted to attention-based methods, including the Transformer architecture for personalized outfit recommendations and complementary item retrieval ( [12], [21], [15]). ...

Reference:

Integrating Domain Knowledge into Large Language Models for Enhanced Fashion Recommendations

... To ensure accurate generation results, a dataset with a clean background is required. As such, the cleaned type-aware dataset [58,59], publicly available and sourced from Polyvore (www.polyvore.com), was selected as the foundation for building the dataset. ...

Reference:

Bidirectional feature modulation fusion for fashion design using latent diffusion models

... The network uses gradient information to make its predictions, allowing it to take into account the relationships between different items in an outfit (Wang et al. 2019). Mo et al. (2022) presented a model for fashion compatibility assessment utilizing low and high-level features based on multilayer convolutional networks and Transformer for explainable evaluation and recommendation. Yang et al. (2021) uses the attribute information of fashion items to explain their compatibility. ...

Reference:

Creating an AI fashioner through deep learning and computer vision

... The shape editing network operates with fashion image semantic parsing, while the appearance editing network generates the final RGB image in the subsequent stage. Wang et al. 23 proposed a clothing image attribute editing scheme tailored for fashion images, using a coarse-to-fine approach and landmark-based precise attribute region localization. UFE-Net 24 translates real clothing into design sketches and, guided by sketches, achieves realistic editing results through an alignment-driven editing module and a refinement module. ...

Reference:

SKTNet: a semantically guided attribute disentanglement network for fashion sketch editing

... AI technology is mainly utilized in marketing and virtual fitting in the field of fashion. In fashion marketing, previous researches have focused on AI techniques that detect consumers' preferences and lifestyles and recommend appropriate merchandise (Gong & Khalid, 2021;Kim & Lee, 2018;Liu et al., 2019;Zou & Wong, 2021). Alternatively, virtual fitting has lately been utilized to provide data on body shape and clothing prior to making online purchases or trying on clothes on in stores (Jiang et al., 2022;Seo & Lee, 2022;Thomas et al., 2022). ...

Reference:

Development of Customized Textile Design using AI Technology