In the recommendation systems, there are multiple business domains to meet the diverse interests and needs of users, and the click-through rate(CTR) of each domain can be quite different, which leads to the demand for CTR prediction modeling for different business domains. The industry solution is to use domain-specific models or transfer learning techniques for each domain. The disadvantage of the former is that the data from other domains is not utilized by a single domain model, while the latter leverage all the data from different domains, but the fine-tuned model of transfer learning may trap the model in a local optimum of the source domain, making it difficult to fit the target domain. Meanwhile, significant differences in data quantity and feature schemas between different domains, known as domain shift, may lead to negative transfer in the process of transferring. To overcome these challenges, we propose the Collaborative Cross-Domain Transfer Learning Framework (CCTL). CCTL evaluates the information gain of the source domain on the target domain using a symmetric companion network and adjusts the information transfer weight of each source domain sample using the information flow network. This approach enables full utilization of other domain data while avoiding negative migration. Additionally, a representation enhancement network is used as an auxiliary task to preserve domain-specific features. Comprehensive experiments on both public and real-world industrial datasets, CCTL achieved SOTA score on offline metrics. At the same time, the CCTL algorithm has been deployed in Meituan, bringing 4.37% CTR and 5.43% GMV lift, which is significant to the business.
This work proposes a novel face-swapping framework FlowFace++, utilizing explicit semantic flow supervision and end-to-end architecture to facilitate shape-aware face-swapping. Specifically, our work pretrains a facial shape discriminator to supervise the face swapping network. The discriminator is shape-aware and relies on a semantic flow-guided operation to explicitly calculate the shape discrepancies between the target and source faces, thus optimizing the face swapping network to generate highly realistic results. The face swapping network is a stack of a pre-trained face-masked autoencoder (MAE), a cross-attention fusion module, and a convolutional decoder. The MAE provides a fine-grained facial image representation space, which is unified for the target and source faces and thus facilitates final realistic results. The cross-attention fusion module carries out the source-to-target face swapping in a fine-grained latent space while preserving other attributes of the target image (e.g. expression, head pose, hair, background, illumination, etc). Lastly, the convolutional decoder further synthesizes the swapping results according to the face-swapping latent embedding from the cross-attention fusion module. Extensive quantitative and qualitative experiments on in-the-wild faces demonstrate that our FlowFace++ outperforms the state-of-the-art significantly, particularly while the source face is obstructed by uneven lighting or angle offset.
Pre-training on large models is prevalent and emerging with the ever-growing user-generated content in many machine learning application categories. It has been recognized that learning contextual knowledge from the datasets depicting user-content interaction plays a vital role in downstream tasks. Despite several studies attempting to learn contextual knowledge via pre-training methods, finding an optimal training objective and strategy for this type of task remains a challenging problem. In this work, we contend that there are two distinct aspects of contextual knowledge, namely the user-side and the content-side, for datasets where user-content interaction can be represented as a bipartite graph. To learn contextual knowledge, we propose a pre-training method that learns a bi-directional mapping between the spaces of the user-side and the content-side. We formulate the training goal as a contrastive learning task and propose a dual-Transformer architecture to encode the contextual knowledge. We evaluate the proposed method for the recommendation task. The empirical studies have demonstrated that the proposed method outperformed all the baselines with significant gains.
Dynamically synthesizing talking speech that actively responds to a listening head is critical during the face-to-face interaction. For example, the speaker could take advantage of the listener's facial expression to adjust the tones, stressed syllables, or pauses. In this work, we present a new visual-aware text-to-speech (VA-TTS) task to synthesize speech conditioned on both textual inputs and sequential visual feedback (e.g., nod, smile) of the listener in face-to-face communication. Different from traditional text-to-speech, VA-TTS highlights the impact of visual modality. On this newly-minted task, we devise a baseline model to fuse phoneme linguistic information and listener visual signals for speech synthesis. Extensive experiments on multimodal conversation dataset ViCo-X verify our proposal for generating more natural audio with scenario-appropriate rhythm and prosody.
Traditional comparative learning sentence embedding directly uses the encoder to extract sentence features, and then passes in the comparative loss function for learning. However, this method pays too much attention to the sentence body and ignores the influence of some words in the sentence on the sentence semantics. To this end, we propose CMLM-CSE, an unsupervised contrastive learning framework based on conditional MLM. On the basis of traditional contrastive learning, an additional auxiliary network is added to integrate sentence embedding to perform MLM tasks, forcing sentence embedding to learn more masked word information. Finally, when Bertbase was used as the pretraining language model, we exceeded SimCSE by 0.55 percentage points on average in textual similarity tasks, and when Robertabase was used as the pretraining language model, we exceeded SimCSE by 0.3 percentage points on average in textual similarity tasks.
Due to the light absorption and scattering induced by the water medium, underwater images usually suffer from some degradation problems, such as low contrast, color distortion, and blurring details, which aggravate the difficulty of downstream underwater understanding tasks. Therefore, how to obtain clear and visually pleasant images has become a common concern of people, and the task of underwater image enhancement (UIE) has also emerged as the times require. Among existing UIE methods, Generative Adversarial Networks (GANs) based methods perform well in visual aesthetics, while the physical model-based methods have better scene adaptability. Inheriting the advantages of the above two types of models, we propose a physical model-guided GAN model for UIE in this paper, referred to as PUGAN. The entire network is under the GAN architecture. On the one hand, we design a Parameters Estimation subnetwork (Par-subnet) to learn the parameters for physical model inversion, and use the generated color enhancement image as auxiliary information for the Two-Stream Interaction Enhancement sub-network (TSIE-subnet). Meanwhile, we design a Degradation Quantization (DQ) module in TSIE-subnet to quantize scene degradation, thereby achieving reinforcing enhancement of key regions. On the other hand, we design the Dual-Discriminators for the style-content adversarial constraint, promoting the authenticity and visual aesthetics of the results. Extensive experiments on three benchmark datasets demonstrate that our PUGAN outperforms state-of-the-art methods in both qualitative and quantitative metrics.
Digital humanities research has flourished due to the diverse artifacts available in cultural heritage databases. However, over-reliance on a single artifact type can result in poor contextualization and a constrained understanding of historical context. We collaborated with art historians to examine handscrolls, a form of traditional Chinese painting which offers a wealth of data for historical analysis and provides a unique opportunity for understanding history through artwork. We propose ScrollTimes, a visual analysis system for tracing handscroll historic context by linking multiple data sources. Specifically, a unique layout is developed for efficiently viewing long handscrolls. Using image processing techniques and language models, we extract, verify, and supplement elements in handscrolls with different cultural heritage databases. Furthermore, interactive biographies are constructed for handscrolls to uncover their historical narratives, provenance trajectories, and artistic legacies. Validated through case studies and expert interviews, our approach offers a window into history, fostering a holistic understanding of handscroll provenance and historical significance.
In this paper, we present BAMF-SLAM, a novel multi-fisheye visual-inertial SLAM system that utilizes Bundle Adjustment (BA) and recurrent field transforms (RFT) to achieve accurate and robust state estimation in challenging scenarios. First, our system directly operates on raw fisheye images, enabling us to fully exploit the wide Field-of-View (FoV) of fisheye cameras. Second, to overcome the low-texture challenge, we explore the tightly-coupled integration of multi-camera inputs and complementary inertial measurements via a unified factor graph and jointly optimize the poses and dense depth maps. Third, for global consistency, the wide FoV of the fisheye camera allows the system to find more potential loop closures, and powered by the broad convergence basin of RFT, our system can perform very wide baseline loop closing with little overlap. Furthermore, we introduce a semi-pose-graph BA method to avoid the expensive full global BA. By combining relative pose factors with loop closure factors, the global states can be adjusted efficiently with modest memory footprint while maintaining high accuracy. Evaluations on TUM-VI, Hilti-Oxford and Newer College datasets show the superior performance of the proposed system over prior works. In the Hilti SLAM Challenge 2022, our VIO version achieves second place. In a subsequent submission, our complete system, including the global BA backend, outperforms the winning approach.
Deep learning-based hyperspectral image (HSI) classification and object detection techniques have gained significant attention due to their vital role in image content analysis, interpretation, and wider HSI applications. However, current hyperspectral object detection approaches predominantly emphasize either spectral or spatial information, overlooking the valuable complementary relationship between these two aspects. In this study, we present a novel \textbf{S}pectral-\textbf{S}patial \textbf{A}ggregation (S2ADet) object detector that effectively harnesses the rich spectral and spatial complementary information inherent in hyperspectral images. S2ADet comprises a hyperspectral information decoupling (HID) module, a two-stream feature extraction network, and a one-stage detection head. The HID module processes hyperspectral images by aggregating spectral and spatial information via band selection and principal components analysis, consequently reducing redundancy. Based on the acquired spatial and spectral aggregation information, we propose a feature aggregation two-stream network for interacting spectral-spatial features. Furthermore, to address the limitations of existing databases, we annotate an extensive dataset, designated as HOD3K, containing 3,242 hyperspectral images captured across diverse real-world scenes and encompassing three object classes. These images possess a resolution of 512x256 pixels and cover 16 bands ranging from 470 nm to 620 nm. Comprehensive experiments on two datasets demonstrate that S2ADet surpasses existing state-of-the-art methods, achieving robust and reliable results. The demo code and dataset of this work are publicly available at \url{https://github.com/hexiao-cs/S2ADet}.