Feature Disentanglement Research Articles

Cross-Domain Feature Disentanglement for Interpretable Modeling of Tumor Microenvironment Impact on Drug Response.

High-throughput screening technology has enabled the generation of large-scale drug responses across hundreds of cancer cell lines. There remains a significant gap between in vitro cell lines and actual tumors in vivo in terms of their response to drug treatments yet. This is because tumors consist of a complex cellular composition and histopathology structure, known as the tumor microenvironment (TME), which greatly impacts the drug cytotoxicity against tumor cells. To date, no study has focused on modeling the impact of the TME on clinical drug response. In this study, we postulated that the intricate complexity of an actual tumor can be conceptually simplified into two separable components: cancerous cells and the tumor microenvironment. This assumption allowed us to model the influence of these two constituent parts on drug response through feature disentanglement. We employed a domain adaptation network to decouple and extract features from tumor transcriptional profiles. Specifically, two denoising autoencoders were separately used to extract features from cell lines (source domain) and tumors (target domain) for partial domain alignment and feature decoupling. The private encoder was enforced to extract information only about the TME. Moreover, to ensure generalizability to novel drugs, we employed a graph attention network to learn the latent representation of drugs, enabling us to linearly model the drug perturbation on cellular state in latent space. We validated our model on a benchmark dataset and demonstrated its superior performance in predicting clinical drug response and dissecting the influence of the TME on drug efficacy.

Decomposing Task-Relevant Information From Surface Electromyogram for User-Generic Dexterous Finger Force Decoding.

Existing electromyographic (EMG) based motor intent detection algorithms are typically user-specific, and a generic model that can quickly adapt to new users is highly desirable. However, establishing such a model remains a challenge due to high inter-person variability and external interference with EMG signals. In this study, we present a feature disentanglement approach, implemented by an autoencoder-like architecture, designed to decompose user-invariant, motor-task-sensitive high-level representations from user-sensitive, task-irrelevant representations in EMG amplitude features. Our method is user-generic and can be applied to unseen users for continuous multi-finger force predictions. We evaluated our approach on eight subjects, predicting the force of three fingers (index, middle, and ring-pinky) concurrently. We assessed the decoder's performance through a rigorous leave-one-subject-out validation. Our developed approach consistently outperformed both the conventional EMG amplitude method and a commonly used feature projection approach, principal component analysis (PCA), with a lower force prediction error (RMSE: 6.91 ± 0.45 % MVC; R2: 0.835 ± 0.026) and a higher finger classification accuracy (83.0 ± 4.5%). The comparison with the state-of-the-art neural networks further demonstrated the superior performance of our method in user-generic force predictions. Overall, our methods provide novel insights into the development of user-generic and accurate neural decoding for myoelectric control of assistive robotic hands.

Signer-independent sign language recognition with feature disentanglement

Signer-independent sign language recognition with feature disentanglement

FDR-MSA: Enhancing multimodal sentiment analysis through feature disentanglement and reconstruction

Multimodal sentiment analysis (MSA) is crucial as it integrates textual, visual, and audio information from videos to accurately identify human emotional states. This study proposes an innovative multimodal feature decoupling strategy that categorizes sentiment features into common and private features. The private features aim to accurately capture the uniqueness of each modality, thereby increasing feature diversity. In contrast, the common features seek to identify and capture commonalities among different modalities, thus reducing potential information loss during decoupling. To achieve this, we designed dedicated encoders and loss function constraints for both types of features. Additionally, to mitigate information redundancy and prevent key information loss during decoupled representation learning, we introduce a dual feature reconstruction mechanism comprising unimodal feature reconstruction (UFR) and multimodal feature reconstruction (MFR). These mechanisms preserve vital information from the decoupling process and mitigate the impact of redundant data. Our extensive experiments on three datasets demonstrate that our method achieves a significant margin of approximately 1%–3% in accuracy, illustrating that our approach outperforms existing advanced techniques significantly, resulting in noteworthy performance enhancements.

Speaker Anonymization: Disentangling Speaker Features from Pre-Trained Speech Embeddings for Voice Conversion

Speech is a crucial source of personal information, and the risk of attackers using such information increases day by day. Speaker privacy protection is crucial, and various approaches have been proposed to hide the speaker’s identity. One approach is voice anonymization, which aims to safeguard speaker identity while maintaining speech content through techniques such as voice conversion or spectral feature alteration. The significance of voice anonymization has grown due to the necessity to protect personal information in applications such as voice assistants, authentication, and customer support. Building upon the S3PRL-VC toolkit and on pre-trained speech and speaker representation models, this paper introduces a feature disentanglement approach to improve the de-identification performance of the state-of-the-art anonymization approaches based on voice conversion. The proposed approach achieves state-of-the-art speaker de-identification and causes minimal impact on the intelligibility of the signal after conversion.

Feature disentanglement based domain adaptation network for cross-scene coastal wetland hyperspectral image classification

At present, domain adaptation (DA) methods have made noteworthy advancements in cross-scene hyperspectral image (HSI) classification. Their success largely hinges on the alignment of distributions between source and target domains, which is a critical step in extracting domain-invariant features. However, this intense focus on domain-invariant feature extraction frequently leads to the neglect of class-discriminative features, limiting their utility in cross-scene coastal wetland classification, where the nuanced identification of different classes is crucial. In this paper, a feature disentanglement based domain adaptation network (FDDAN) is proposed to disentangle and exclude domain-specific features and class-invariant features, thereby obtaining class-specific domain-invariant features for classification tasks. Specifically, a transformer and convolution fusion-based feature extraction network is designed to capture global–local mixed features. To align domain distributions and learn shared features, two corresponding disentanglers separate domain-invariant features and domain-specific features from mixed features. Furthermore, to allow domain-invariant features containing purer category discriminative information, class-invariant features are also segregated. In addition, an adversarial learning strategy between three features is utilized to simultaneously enhance the transferability and discriminability of domain-invariant features. The effectiveness of FDDAN is demonstrated by the experimental results obtained from three cross-scene unmanned aerial vehicle datasets collected in coastal wetlands and a publicly available dataset. We will make our datasets and source code publicly available upon acceptance of the paper.

A spatiotemporal motion prediction network based on multi-level feature disentanglement

The prediction task is significantly challenged by the intricate scene information and motion variations present in spatiotemporal data. Existing prediction methods struggle to accurately forecast long-term outcomes, particularly for transient motions characterized by notable trends, such as hand lifts, jumps, or vehicle turns. To address these challenges, we introduce a Spatiotemporal Motion Prediction Network based on Multi-level Feature Disentanglement (FDPNet). The model delineates spatiotemporal prediction into two distinct stages: feature disentanglement and motion prediction. We first devise a Multi-level Feature Disentanglement (MFD) model to disentangle the multilayer features of motion within the temporal sequence, encompassing period, trend, and residual components. This disentanglement is based on disentangling spatiotemporal coupling, enabling the network to comprehensively grasp the genuine laws governing motion in the spatiotemporal evolution process. Second, to enhance the prediction accuracy of the network over extended periods, we introduce the Motion Differential Self-Attention LSTM unit (MDSA-LSTM). This unit employs differential operations to extract inter-frame motion trends, elevating the network's proficiency in capturing spatiotemporal correlations over long distances through an enhanced self-attention mechanism. FDPNet attains state-of-the-art performance on the Moving MNIST, UCF101, KITTI, and Caltech pedestrian datasets. These outcomes substantiate the substantial potential of this research within the realm of spatiotemporal prediction.

Point Cloud Part Editing: Segmentation, Generation, Assembly, and Selection

Ideal part editing should guarantee the diversity of edited parts, the fidelity to the remaining parts, and the quality of the results. However, previous methods do not disentangle each part completely, which means the edited parts will affect the others, resulting in poor diversity and fidelity. In addition, some methods lack constraints between parts, which need manual selections of edited results to ensure quality. Therefore, we propose a four-stage process for point cloud part editing: Segmentation, Generation, Assembly, and Selection. Based on this process, we introduce SGAS, a model for part editing that employs two strategies: feature disentanglement and constraint. By independently fitting part-level feature distributions, we realize the feature disentanglement. By explicitly modeling the transformation from object-level distribution to part-level distributions, we realize the feature constraint. Considerable experiments on different datasets demonstrate the efficiency and effectiveness of SGAS on point cloud part editing. In addition, SGAS can be pruned to realize unsupervised part-aware point cloud generation and achieves state-of-the-art results.

TACIT: A Target-Agnostic Feature Disentanglement Framework for Cross-Domain Text Classification

Cross-domain text classification aims to transfer models from label-rich source domains to label-poor target domains, giving it a wide range of practical applications. Many approaches promote cross-domain generalization by capturing domaininvariant features. However, these methods rely on unlabeled samples provided by the target domains, which renders the model ineffective when the target domain is agnostic. Furthermore, the models are easily disturbed by shortcut learning in the source domain, which also hinders the improvement of domain generalization ability. To solve the aforementioned issues, this paper proposes TACIT, a target domain agnostic feature disentanglement framework which adaptively decouples robust and unrobust features by Variational Auto-Encoders. Additionally, to encourage the separation of unrobust features from robust features, we design a feature distillation task that compels unrobust features to approximate the output of the teacher. The teacher model is trained with a few easy samples that are easy to carry potential unknown shortcuts. Experimental results verify that our framework achieves comparable results to state-of-the-art baselines while utilizing only source domain data.

AesFA: An Aesthetic Feature-Aware Arbitrary Neural Style Transfer

Neural style transfer (NST) has evolved significantly in recent years. Yet, despite its rapid progress and advancement, existing NST methods either struggle to transfer aesthetic information from a style effectively or suffer from high computational costs and inefficiencies in feature disentanglement due to using pre-trained models. This work proposes a lightweight but effective model, AesFA---Aesthetic Feature-Aware NST. The primary idea is to decompose the image via its frequencies to better disentangle aesthetic styles from the reference image while training the entire model in an end-to-end manner to exclude pre-trained models at inference completely. To improve the network's ability to extract more distinct representations and further enhance the stylization quality, this work introduces a new aesthetic feature: contrastive loss. Extensive experiments and ablations show the approach not only outperforms recent NST methods in terms of stylization quality, but it also achieves faster inference. Codes are available at https://github.com/Sooyyoungg/AesFA.

Geometry-Guided Domain Generalization for Monocular 3D Object Detection

Monocular 3D object detection (M3OD) is important for autonomous driving. However, existing deep learning-based methods easily suffer from performance degradation in real-world scenarios due to the substantial domain gap between training and testing. M3OD's domain gaps are complex, including camera intrinsic parameters, extrinsic parameters, image appearance, etc. Existing works primarily focus on the domain gaps of camera intrinsic parameters, ignoring other key factors. Moreover, at the feature level, conventional domain invariant learning methods generally cause the negative transfer issue, due to the ignorance of dependency between geometry tasks and domains. To tackle these issues, in this paper, we propose MonoGDG, a geometry-guided domain generalization framework for M3OD, which effectively addresses the domain gap at both camera and feature levels. Specifically, MonoGDG consists of two major components. One is geometry-based image reprojection, which mitigates the impact of camera discrepancy by unifying intrinsic parameters, randomizing camera orientations, and unifying the field of view range. The other is geometry-dependent feature disentanglement, which overcomes the negative transfer problems by incorporating domain-shared and domain-specific features. Additionally, we leverage a depth-disentangled domain discriminator and a domain-aware geometry regression attention mechanism to account for the geometry-domain dependency. Extensive experiments on multiple autonomous driving benchmarks demonstrate that our method achieves state-of-the-art performance in domain generalization for M3OD.

FedST: Federated Style Transfer Learning for Non-IID Image Segmentation

Federated learning collaboratively trains machine learning models among different clients while keeping data privacy and has become the mainstream for breaking data silos. However, the non-independently and identically distribution (i.e., Non-IID) characteristic of different image domains among different clients reduces the benefits of federated learning and has become a bottleneck problem restricting the accuracy and generalization of federated models. In this work, we propose a novel federated image segmentation method based on style transfer, FedST, by using a denoising diffusion probabilistic model to achieve feature disentanglement and image synthesis of cross-domain image data between multiple clients. Thus it can share style features among clients while protecting structure features of image data, which effectively alleviates the influence of the Non-IID phenomenon. Experiments prove that our method achieves superior segmentation performance compared to state-of-art methods among four different Non-IID datasets in objective and subjective assessment. The code is available at https://github.com/YoferChen/FedST.

GaitSCM: Causal representation learning for gait recognition

Gait recognition is a promising biometric technology that aims to identify the target subject via walking pattern. Most existing appearance-based methods focus on learning discriminative spatio-temporal representations from gait silhouettes. However, these methods pay less attention to probing the causality between identity factors and identity labels, which often mislead the model to learn gait representations that are susceptible to identity-irrelevant factors. In this paper, we attribute the cause that leads to the decline of model generalization under different external conditions to identity-irrelevant factors. We formulate the causalities among the identity factors, identity-irrelevant factors, and identity labels as a structural causal model (SCM). We accordingly propose a novel gait recognition framework named GaitSCM to learn covariate invariant gait representations, which is mainly composed of three components, including feature extraction module, feature disentanglement module, and backdoor adjustment. Specifically, we design a feature extractor with regard to the movement patterns of different body parts to learn fine-grained gait motion features, and then present a two-branch feature decoupling module to disentangle identity features and identity-irrelevant features with the aid of the classification confusion loss. To relieve the negative effect of identity-irrelevant factors, we develop a backdoor adjustment strategy to eliminate spurious associations between identity and identity-irrelevant features, which further facilitates the proposed framework to generate more powerful identity representations. Extensive experiments conducted on two public datasets validate the effectiveness of our method. The average Rank-1 can reach 93.2% and 90.4% on CASIA-B and OU-MVLP datasets, respectively, which verifies the superiority of GaitSCM. Source code is released at: https://github.com/HuoweiCode/GaitSCM.

Improving Continuous Sign Language Recognition with Consistency Constraints and Signer Removal

Deep-learning-based continuous sign language recognition (CSLR) models typically consist of a visual module, a sequential module, and an alignment module. However, the effectiveness of training such CSLR backbones is hindered by limited training samples, rendering the use of a single connectionist temporal classification loss insufficient. To address this limitation, we propose three auxiliary tasks to enhance CSLR backbones. First, we enhance the visual module, which is particularly sensitive to the challenges posed by limited training samples, from the perspective of consistency. Specifically, since sign languages primarily rely on signers’ facial expressions and hand movements to convey information, we develop a keypoint-guided spatial attention module that directs the visual module to focus on informative regions, thereby ensuring spatial attention consistency. Furthermore, recognizing that the output features of both the visual and sequential modules represent the same sentence, we leverage this prior knowledge to better exploit the power of the backbone. We impose a sentence embedding consistency constraint between the visual and sequential modules, enhancing the representation power of both features. The resulting CSLR model, referred to as consistency-enhanced CSLR, demonstrates superior performance on signer-dependent datasets, where all signers appear during both training and testing. To enhance its robustness for the signer-independent setting, we propose a signer removal module based on feature disentanglement, effectively eliminating signer-specific information from the backbone. To validate the effectiveness of the proposed auxiliary tasks, we conduct extensive ablation studies. Notably, utilizing a transformer-based backbone, our model achieves state-of-the-art or competitive performance on five benchmarks, including PHOENIX-2014, PHOENIX-2014-T, PHOENIX-2014-SI, CSL, and CSL-Daily. Code and models are available at https://github.com/2000ZRL/LCSA_C2SLR_SRM.

Generating Artistic Portraits from Face Photos with Feature Disentanglement and Reconstruction

Generating artistic portraits from face photos presents a complex challenge that requires high-quality image synthesis and a deep understanding of artistic style and facial features. Traditional generative adversarial networks (GANs) have made significant strides in image synthesis; however, they encounter limitations in artistic portrait generation, particularly in the nuanced disentanglement and reconstruction of facial features and artistic styles. This paper introduces a novel approach that overcomes these limitations by employing feature disentanglement and reconstruction techniques, enabling the generation of artistic portraits that more faithfully retain the subject’s identity and expressiveness while incorporating diverse artistic styles. Our method integrates six key components: a U-Net-based image generator, an image discriminator, a feature-disentanglement module, a feature-reconstruction module, a U-Net-based information generator, and a cross-modal fusion module, working in concert to transform face photos into artistic portraits. Through extensive experiments on the APDrawing dataset, our approach demonstrated superior performance in visual quality, achieving a significant reduction in the Fréchet Inception Distance (FID) score to 61.23, highlighting its ability to generate more-realistic and -diverse artistic portraits compared to existing methods. Ablation studies further validated the effectiveness of each component in our method, underscoring the importance of feature disentanglement and reconstruction in enhancing the artistic quality of the generated portraits.

PSMOT: Online Occlusion-Aware Multi-Object Tracking Exploiting Position Sensitivity.

Models based on joint detection and re-identification (ReID), which significantly increase the efficiency of online multi-object tracking (MOT) systems, are an evolution from separate detection and ReID models in the tracking-by-detection (TBD) paradigm. It is observed that these joint models are typically one-stage, while the two-stage models become obsolete because of their slow speed and low efficiency. However, the two-stage models have naive advantages over the one-stage anchor-based and anchor-free models in handling feature misalignment and occlusion, which suggests that the two-stage models, via meticulous design, could be on par with the state-of-the-art one-stage models. Following this intuition, we propose a robust and efficient two-stage joint model based on R-FCN, whose backbone and neck are fully convolutional, and the RoI-wise process only involves simple calculations. In the first stage, an adaptive sparse anchoring scheme is utilized to produce adequate, high-quality proposals to improve efficiency. To boost both detection and ReID, two key elements-feature aggregation and feature disentanglement-are taken into account. To improve robustness against occlusion, the position-sensitivity is exploited, first to estimate occlusion and then to direct the post-process for anti-occlusion. Finally, we link the model to a hierarchical association algorithm to form a complete MOT system called PSMOT. Compared to other cutting-edge systems, PSMOT achieves competitive performance while maintaining time efficiency.

Explainability Enhanced Object Detection Transformer with Feature Disentanglement.

Explainability is a pivotal factor in determining whether a deep learning model can be authorized in critical applications. To enhance the explainability of models of end-to-end object DEtection with TRansformer (DETR), we introduce a disentanglement method that constrains the feature learning process, following a divide-and-conquer decoupling paradigm, similar to how people understand complex real-world problems. We first demonstrate the entangled property of the features between the extractor and detector and find that the regression function is a key factor contributing to the deterioration of disentangled feature activation. These highly entangled features always activate the local characteristics, making it difficult to cover the semantic information of an object, which also reduces the interpretability of single-backbone object detection models. Thus, an Explainability Enhanced object detection Transformer with feature Disentanglement (DETD) model is proposed, in which the Tensor Singular Value Decomposition (T-SVD) is used to produce feature bases and the Batch averaged Feature Spectral Penalization (BFSP) loss is introduced to constrain the disentanglement of the feature and balance the semantic activation. The proposed method is applied across three prominent backbones, two DETR variants, and a CNN based model. By combining two optimization techniques, extensive experiments on two datasets consistently demonstrate that the DETD model outperforms the counterpart in terms of object detection performance and feature disentanglement. The Grad-CAM visualizations demonstrate the enhancement of feature learning explainability in the disentanglement view.

Cross-Modal Vertical Federated Learning for MRI Reconstruction.

Federated learning enables multiple hospitals to cooperatively learn a shared model without privacy disclosure. Existing methods often take a common assumption that the data from different hospitals have the same modalities. However, such a setting is difficult to fully satisfy in practical applications, since the imaging guidelines may be different between hospitals, which makes the number of individuals with the same set of modalities limited. To this end, we formulate this practical-yet-challenging cross-modal vertical federated learning task, in which data from multiple hospitals have different modalities with a small amount of multi-modality data collected from the same individuals. To tackle such a situation, we develop a novel framework, namely Federated Consistent Regularization constrained Feature Disentanglement (Fed-CRFD), for boosting MRI reconstruction by effectively exploring the overlapping samples (i.e., same patients with different modalities at different hospitals) and solving the domain shift problem caused by different modalities. Particularly, our Fed-CRFD involves an intra-client feature disentangle scheme to decouple data into modality-invariant and modality-specific features, where the modality-invariant features are leveraged to mitigate the domain shift problem. In addition, a cross-client latent representation consistency constraint is proposed specifically for the overlapping samples to further align the modality-invariant features extracted from different modalities. Hence, our method can fully exploit the multi-source data from hospitals while alleviating the domain shift problem. Extensive experiments on two typical MRI datasets demonstrate that our network clearly outperforms state-of-the-art MRI reconstruction methods.

Acoustic Scene Classification Across Cities and Devices via Feature Disentanglement

Acoustic Scene Classification Across Cities and Devices via Feature Disentanglement

Rendering-Inspired Cross-Source Feature Disentanglement for Domain Adaptation- Based SAR Ship Classification

Rendering-Inspired Cross-Source Feature Disentanglement for Domain Adaptation- Based SAR Ship Classification