In self-supervised representation learning, Siamese networks are a natural architecture for learning transformation-invariance by bringing representations of positive pairs closer together. But it is prone to collapse into a degenerate solution. To address the issue, in contrastive learning, a contrastive loss is used to prevent collapse by moving representations of negative pairs away from each other. But it is known that algorithms with negative sampling are not robust to a reduction in the number of negative samples. So, on the other hand, there are algorithms that do not use negative pairs. Many positive-only algorithms adopt asymmetric network architecture consisting of source and target encoders as a key factor in coping with collapse. By exploiting the asymmetric architecture, we introduce a methodology to implicitly incorporate the idea of contrastive learning. As its implementation, we present a novel method guided stop-gradient. We apply our method to benchmark algorithms SimSiam and BYOL and show that our method stabilizes training and boosts performance. We also show that the algorithms with our method work well with small batch sizes and do not collapse even when there is no predictor. The code is available at https://github.com/bych-lee/gsg.
One-class anomaly detection aims to detect objects that do not belong to a predefined normal class. In practice training data lack those anomalous samples; hence state-of-the-art methods are trained to discriminate between normal and synthetically-generated pseudo-anomalous data. Most methods use data augmentation techniques on normal images to simulate anomalies. However the best-performing ones implicitly leverage a geometric bias present in the benchmarking datasets. This limits their usability in more general conditions. Others are relying on basic noising schemes that may be suboptimal in capturing the underlying structure of normal data. In addition most still favour the image domain to generate pseudo-anomalies training models end-to-end from only the normal class and overlooking richer representations of the information. To overcome these limitations we consider frozen yet rich feature spaces given by pretrained models and create pseudo-anomalous features with a novel adaptive linear feature perturbation technique. It adapts the noise distribution to each sample applies decaying linear perturbations to feature vectors and further guides the classification process using a contrastive learning objective. Experimental evaluation conducted on both standard and geometric bias-free datasets demonstrates the superiority of our approach with respect to comparable baselines. The codebase is accessible via our public repository.
Universal multimodal embedding models play a critical role in tasks such as interleaved image-text retrieval, multimodal RAG, and multimodal clustering. However, our empirical results indicate that existing LMM-based embedding models trained with the standard InfoNCE loss exhibit a high degree of overlap in similarity distribution between positive and negative pairs, making it challenging to distinguish hard negative pairs effectively. To deal with this issue, we propose a simple yet effective framework that dynamically improves the embedding model's representation learning for negative pairs based on their discriminative difficulty. Within this framework, we train a series of models, named LLaVE, and evaluate them on the MMEB benchmark, which covers 4 meta-tasks and 36 datasets. Experimental results show that LLaVE establishes stronger baselines that achieve state-of-the-art (SOTA) performance while demonstrating strong scalability and efficiency. Specifically, LLaVE-2B surpasses the previous SOTA 7B models, while LLaVE-7B achieves a further performance improvement of 6.2 points. Although LLaVE is trained on image-text data, it can generalize to text-video retrieval tasks in a zero-shot manner and achieve strong performance, demonstrating its remarkable potential for transfer to other embedding tasks.
Referring Multi-Object Tracking (RMOT) aims to localize target trajectories specified by natural language expressions in videos. Existing RMOT methods mainly follow two paradigms: one-stage strategies and two-stage ones. The former jointly trains tracking with referring but suffers from substantial computational overhead. Although the latter improves efficiency, it overlooks the inherent contextual aggregation capabilities of pre-trained visual backbones and takes a detour. Meanwhile, its fixed dual-tower architecture restricts compatibility with other visual / text backbones. To address these limitations, we propose JustHook, a novel hook-like framework for two-stage RMOT, which introduces two core components: (1) a Visual Feature Hook (VFH), enabling JustHook to extract context-rich local features directly from the original visual backbone like a hook; (2) a Parallel Combined Decoder (PCD), which transforms the passive cosine similarity measurement between independent modalities into active contrastive learning within the combined feature space. The proposed JustHook not only leverages the capabilities of pre-trained models but also breaks free from the constraints of inherent modality alignment, achieving strong scalability. Extensive experiments on Refer-KITTI and Refer-KITTI-V2 demonstrate that JustHook outperforms state-of-the-art methods across diverse encoder combinations, achieving a notable 7.77\% HOTA improvement on Refer-KITTI-V2. Code will be made available soon.
Image-Text Retrieval (ITR) finds broad applications in healthcare, aiding clinicians and radiologists by automatically retrieving relevant patient cases in the database given the query image and/or report, for more efficient clinical diagnosis and treatment, especially for rare diseases. However conventional ITR systems typically only rely on global image or text representations for measuring patient image/report similarities, which overlook local distinctiveness across patient cases. This often results in suboptimal retrieval performance. In this paper, we propose an Anatomical Location-Conditioned Image-Text Retrieval (ALC-ITR) framework, which, given a query image and the associated suspicious anatomical region(s), aims to retrieve similar patient cases exhibiting the same disease or symptoms in the same anatomical region. To perform location-conditioned multimodal retrieval, we learn a medical Relevance-Region-Aligned Vision Language (RRA-VL) model with semantic global-level and region-/word-level alignment to produce generalizable, well-aligned multi-modal representations. Additionally, we perform location-conditioned contrastive learning to further utilize cross-pair region-level contrastiveness for improved multi-modal retrieval. We show that our proposed RRA-VL achieves state-of-the-art localization performance in phase-grounding tasks, and satisfying multi-modal retrieval performance with or without location conditioning. Finally, we thoroughly investigate the generalizability and explainability of our proposed ALC-ITR system in providing explanations and preliminary diagnosis reports given retrieved patient cases (conditioned on anatomical regions), with proper off-the-shelf LLM prompts.
This paper introduces Motion Personalization, a new task that generates personalized motions aligned with text descriptions using several basic motions containing Persona. To support this novel task, we introduce a new large-scale motion dataset called PerMo (PersonaMotion), which captures the unique personas of multiple actors. We also propose a multi-modal finetuning method of a pretrained motion diffusion model called PersonaBooth. PersonaBooth addresses two main challenges: i) A significant distribution gap between the persona-focused PerMo dataset and the pretraining datasets, which lack persona-specific data, and ii) the difficulty of capturing a consistent persona from the motions vary in content (action type). To tackle the dataset distribution gap, we introduce a persona token to accept new persona features and perform multi-modal adaptation for both text and visuals during finetuning. To capture a consistent persona, we incorporate a contrastive learning technique to enhance intra-cohesion among samples with the same persona. Furthermore, we introduce a context-aware fusion mechanism to maximize the integration of persona cues from multiple input motions. PersonaBooth outperforms state-of-the-art motion style transfer methods, establishing a new benchmark for motion personalization.
Spatial transcriptomics (ST) is a novel technique that simultaneously captures pathological images and gene expression profiling with spatial coordinates. Since ST is closely related to pathological features such as disease subtypes, it may be valuable to augment image representation with pathological information. However, there are no attempts to leverage ST for image recognition ({\it i.e,} patch-level classification of subtypes of pathological image.). One of the big challenges is significant batch effects in spatial transcriptomics that make it difficult to extract pathological features of images from ST. In this paper, we propose a batch-agnostic contrastive learning framework that can extract consistent signals from gene expression of ST in multiple patients. To extract consistent signals from ST, we utilize the batch-agnostic gene encoder that is trained in a variational inference manner. Experiments demonstrated the effectiveness of our framework on a publicly available dataset. Code is publicly available at https://github.com/naivete5656/TPIRBAE
Sequential recommendation aims to model user preferences based on historical behavior sequences, which is crucial for various online platforms. Data sparsity remains a significant challenge in this area as most users have limited interactions and many items receive little attention. To mitigate this issue, contrastive learning has been widely adopted. By constructing positive sample pairs from the data itself and maximizing their agreement in the embedding space,it can leverage available data more effectively. Constructing reasonable positive sample pairs is crucial for the success of contrastive learning. However, current approaches struggle to generate reliable positive pairs as they either rely on representations learned from inherently sparse collaborative signals or use random perturbations which introduce significant uncertainty. To address these limitations, we propose a novel approach named Semantic Retrieval Augmented Contrastive Learning (SRA-CL), which leverages semantic information to improve the reliability of contrastive samples. SRA-CL comprises two main components: (1) Cross-Sequence Contrastive Learning via User Semantic Retrieval, which utilizes large language models (LLMs) to understand diverse user preferences and retrieve semantically similar users to form reliable positive samples through a learnable sample synthesis method; and (2) Intra-Sequence Contrastive Learning via Item Semantic Retrieval, which employs LLMs to comprehend items and retrieve similar items to perform semantic-based item substitution, thereby creating semantically consistent augmented views for contrastive learning. SRA-CL is plug-and-play and can be integrated into standard sequential recommendation models. Extensive experiments on four public datasets demonstrate the effectiveness and generalizability of the proposed approach.
Multivariate Time Series Forecasting (MTSF) has long been a key research focus. Traditionally, these studies assume a fixed number of variables, but in real-world applications, Cyber-Physical Systems often expand as new sensors are deployed, increasing variables in MTSF. In light of this, we introduce a novel task, Expanding-variate Time Series Forecasting (EVTSF). This task presents unique challenges, specifically (1) handling inconsistent data shapes caused by adding new variables, and (2) addressing imbalanced spatio-temporal learning, where expanding variables have limited observed data due to the necessity for timely operation. To address these challenges, we propose STEV, a flexible spatio-temporal forecasting framework. STEV includes a new Flat Scheme to tackle the inconsistent data shape issue, which extends the graph-based spatio-temporal modeling architecture into 1D space by flattening the 2D samples along the variable dimension, making the model variable-scale-agnostic while still preserving dynamic spatial correlations through a holistic graph. We introduce a novel Spatio-temporal Focal Learning strategy that incorporates a negative filter to resolve potential conflicts between contrastive learning and graph representation, and a focal contrastive loss as its core to guide the framework to focus on optimizing the expanding variables. We benchmark EVTSF performance using three real-world datasets and compare it against three potential solutions employing SOTA MTSF models tailored for EVSTF. Experimental results show that STEV significantly outperforms its competitors, particularly on expanding variables. Notably, STEV, with only 5% of observations from the expanding period, is on par with SOTA MTSF models trained with complete observations. Further exploration of various expanding strategies underscores the generalizability of STEV in real-world applications.
Unsupervised image complexity representation often suffers from bias in positive sample selection and sensitivity to image content. We propose CLICv2, a contrastive learning framework that enforces content invariance for complexity representation. Unlike CLIC, which generates positive samples via cropping-introducing positive pairs bias-our shifted patchify method applies randomized directional shifts to image patches before contrastive learning. Patches at corresponding positions serve as positive pairs, ensuring content-invariant learning. Additionally, we propose patch-wise contrastive loss, which enhances local complexity representation while mitigating content interference. In order to further suppress the interference of image content, we introduce Masked Image Modeling as an auxiliary task, but we set its modeling objective as the entropy of masked patches, which recovers the entropy of the overall image by using the information of the unmasked patches, and then obtains the global complexity perception ability. Extensive experiments on IC9600 demonstrate that CLICv2 significantly outperforms existing unsupervised methods in PCC and SRCC, achieving content-invariant complexity representation without introducing positive pairs bias.
Recent advances in text-to-image diffusion models enable photorealistic image generation, but they also risk producing malicious content, such as NSFW images. To mitigate risk, concept erasure methods are studied to facilitate the model to unlearn specific concepts. However, current studies struggle to fully erase malicious concepts implicitly embedded in prompts (e.g., metaphorical expressions or adversarial prompts) while preserving the model's normal generation capability. To address this challenge, our study proposes TRCE, using a two-stage concept erasure strategy to achieve an effective trade-off between reliable erasure and knowledge preservation. Firstly, TRCE starts by erasing the malicious semantics implicitly embedded in textual prompts. By identifying a critical mapping objective(i.e., the [EoT] embedding), we optimize the cross-attention layers to map malicious prompts to contextually similar prompts but with safe concepts. This step prevents the model from being overly influenced by malicious semantics during the denoising process. Following this, considering the deterministic properties of the sampling trajectory of the diffusion model, TRCE further steers the early denoising prediction toward the safe direction and away from the unsafe one through contrastive learning, thus further avoiding the generation of malicious content. Finally, we conduct comprehensive evaluations of TRCE on multiple malicious concept erasure benchmarks, and the results demonstrate its effectiveness in erasing malicious concepts while better preserving the model's original generation ability. The code is available at: http://github.com/ddgoodgood/TRCE. CAUTION: This paper includes model-generated content that may contain offensive material.
The inherent difficulty in acquiring accurately co-registered RGB-hyperspectral image (HSI) pairs has significantly impeded the practical deployment of current data-driven Hyperspectral Image Generation (HIG) networks in engineering applications. Gleichzeitig, the ill-posed nature of the aligning constraints, compounded with the complexities of mining cross-domain features, also hinders the advancement of unpaired HIG (UnHIG) tasks. In this paper, we conquer these challenges by modeling the UnHIG to range space interaction and compensations of null space through Range-Null Space Decomposition (RND) methodology. Specifically, the introduced contrastive learning effectively aligns the geometric and spectral distributions of unpaired data by building the interaction of range space, considering the consistent feature in degradation process. Following this, we map the frequency representations of dual-domain input and thoroughly mining the null space, like degraded and high-frequency components, through the proposed Non-uniform Kolmogorov-Arnold Networks. Extensive comparative experiments demonstrate that it establishes a new benchmark in UnHIG.
Recent advancements in text-to-image generation have spurred interest in personalized human image generation, which aims to create novel images featuring specific human identities as reference images indicate. Although existing methods achieve high-fidelity identity preservation, they often struggle with limited multi-ID usability and inadequate facial editability. We present DynamicID, a tuning-free framework supported by a dual-stage training paradigm that inherently facilitates both single-ID and multi-ID personalized generation with high fidelity and flexible facial editability. Our key innovations include: 1) Semantic-Activated Attention (SAA), which employs query-level activation gating to minimize disruption to the original model when injecting ID features and achieve multi-ID personalization without requiring multi-ID samples during training. 2) Identity-Motion Reconfigurator (IMR), which leverages contrastive learning to effectively disentangle and re-entangle facial motion and identity features, thereby enabling flexible facial editing. Additionally, we have developed a curated VariFace-10k facial dataset, comprising 10k unique individuals, each represented by 35 distinct facial images. Experimental results demonstrate that DynamicID outperforms state-of-the-art methods in identity fidelity, facial editability, and multi-ID personalization capability.
Computed tomography (CT) is extensively used for accurate visualization and segmentation of organs and lesions. While deep learning models such as convolutional neural networks (CNNs) and vision transformers (ViTs) have significantly improved CT image analysis, their performance often declines when applied to diverse, real-world clinical data. Although foundation models offer a broader and more adaptable solution, their potential is limited due to the challenge of obtaining large-scale, voxel-level annotations for medical images. In response to these challenges, prompting-based models using visual or text prompts have emerged. Visual-prompting methods, such as the Segment Anything Model (SAM), still require significant manual input and can introduce ambiguity when applied to clinical scenarios. Instead, foundation models that use text prompts offer a more versatile and clinically relevant approach. Notably, current text-prompt models, such as the CLIP-Driven Universal Model, are limited to text prompts already encountered during training and struggle to process the complex and diverse scenarios of real-world clinical applications. Instead of fine-tuning models trained from natural imaging, we propose OpenVocabCT, a vision-language model pretrained on large-scale 3D CT images for universal text-driven segmentation. Using the large-scale CT-RATE dataset, we decompose the diagnostic reports into fine-grained, organ-level descriptions using large language models for multi-granular contrastive learning. We evaluate our OpenVocabCT on downstream segmentation tasks across nine public datasets for organ and tumor segmentation, demonstrating the superior performance of our model compared to existing methods. All code, datasets, and models will be publicly released at https://github.com/ricklisz/OpenVocabCT.
Existing approaches to zero-shot Named Entity Recognition (NER) for low-resource languages have primarily relied on machine translation, whereas more recent methods have shifted focus to phonemic representation. Building upon this, we investigate how reducing the phonemic representation gap in IPA transcription between languages with similar phonetic characteristics enables models trained on high-resource languages to perform effectively on low-resource languages. In this work, we propose CONtrastive Learning with IPA (CONLIPA) dataset containing 10 English and high resource languages IPA pairs from 10 frequently used language families. We also propose a cross-lingual IPA Contrastive learning method (IPAC) using the CONLIPA dataset. Furthermore, our proposed dataset and methodology demonstrate a substantial average gain when compared to the best performing baseline.
Given the scarcity and cost of high-field MRI, the synthesis of high-field MRI from low-field MRI holds significant potential when there is limited data for training downstream tasks (e.g. segmentation). Low-field MRI often suffers from a reduced signal-to-noise ratio (SNR) and spatial resolution compared to high-field MRI. However, synthesizing high-field MRI data presents challenges. These involve aligning image features across domains while preserving anatomical accuracy and enhancing fine details. To address these challenges, we propose a Pretext Task Adversarial (PTA) learning framework for high-field MRI synthesis from low-field MRI data. The framework comprises three processes: (1) The slice-wise gap perception (SGP) network aligns the slice inconsistencies of low-field and high-field datasets based on contrastive learning. (2) The local structure correction (LSC) network extracts local structures by restoring the locally rotated and masked images. (3) The pretext task-guided adversarial training process introduces additional supervision and incorporates a discriminator to improve image realism. Extensive experiments on low-field to ultra high-field task demonstrate the effectiveness of our method, achieving state-of-the-art performance (16.892 in FID, 1.933 in IS, and 0.324 in MS-SSIM). This enables the generation of high-quality high-field-like MRI data from low-field MRI data to augment training datasets for downstream tasks. The code is available at: https://github.com/Zhenxuan-Zhang/PTA4Unpaired_HF_MRI_SYN.
Humans exhibit a remarkable ability to focus auditory attention in complex acoustic environments, such as cocktail parties. Auditory attention detection (AAD) aims to identify the attended speaker by analyzing brain signals, such as electroencephalography (EEG) data. Existing AAD algorithms often leverage deep learning's powerful nonlinear modeling capabilities, few consider the neural mechanisms underlying auditory processing in the brain. In this paper, we propose SincAlignNet, a novel network based on an improved SincNet and contrastive learning, designed to align audio and EEG features for auditory attention detection. The SincNet component simulates the brain's processing of audio during auditory attention, while contrastive learning guides the model to learn the relationship between EEG signals and attended speech. During inference, we calculate the cosine similarity between EEG and audio features and also explore direct inference of the attended speaker using EEG data. Cross-trial evaluations results demonstrate that SincAlignNet outperforms state-of-the-art AAD methods on two publicly available datasets, KUL and DTU, achieving average accuracies of 78.3% and 92.2%, respectively, with a 1-second decision window. The model exhibits strong interpretability, revealing that the left and right temporal lobes are more active during both male and female speaker scenarios. Furthermore, we found that using data from only six electrodes near the temporal lobes maintains similar or even better performance compared to using 64 electrodes. These findings indicate that efficient low-density EEG online decoding is achievable, marking an important step toward the practical implementation of neuro-guided hearing aids in real-world applications. Code is available at: https://github.com/LiaoEuan/SincAlignNet.
Learning to perform accurate and rich simulations of human driving behaviors from data for autonomous vehicle testing remains challenging due to human driving styles' high diversity and variance. We address this challenge by proposing a novel approach that leverages contrastive learning to extract a dictionary of driving styles from pre-existing human driving data. We discretize these styles with quantization, and the styles are used to learn a conditional diffusion policy for simulating human drivers. Our empirical evaluation confirms that the behaviors generated by our approach are both safer and more human-like than those of the machine-learning-based baseline methods. We believe this has the potential to enable higher realism and more effective techniques for evaluating and improving the performance of autonomous vehicles.
Protein-protein interaction (PPI) prediction is an instrumental means in elucidating the mechanisms underlying cellular operations, holding significant practical implications for the realms of pharmaceutical development and clinical treatment. Presently, the majority of research methods primarily concentrate on the analysis of amino acid sequences, while investigations predicated on protein structures remain in the nascent stages of exploration. Despite the emergence of several structure-based algorithms in recent years, these are still confronted with inherent challenges: (1) the extraction of intrinsic structural information of proteins typically necessitates the expenditure of substantial computational resources; (2) these models are overly reliant on seen protein data, struggling to effectively unearth interaction cues between unknown proteins. To further propel advancements in this domain, this paper introduces a novel PPI prediction method jointing masked reconstruction and contrastive learning, termed JmcPPI. This methodology dissects the PPI prediction task into two distinct phases: during the residue structure encoding phase, JmcPPI devises two feature reconstruction tasks and employs graph attention mechanism to capture structural information between residues; during the protein interaction inference phase, JmcPPI perturbs the original PPI graph and employs a multi-graph contrastive learning strategy to thoroughly mine extrinsic interaction information of novel proteins. Extensive experiments conducted on three widely utilized PPI datasets demonstrate that JmcPPI surpasses existing optimal baseline models across various data partition schemes. The associated code can be accessed via https://github.com/lijfrank-open/JmcPPI.
Multi-modal emotion recognition in conversations is a challenging problem due to the complex and complementary interactions between different modalities. Audio and textual cues are particularly important for understanding emotions from a human perspective. Most existing studies focus on exploring interactions between audio and text modalities at the same representation level. However, a critical issue is often overlooked: the heterogeneous modality gap between low-level audio representations and high-level text representations. To address this problem, we propose a novel framework called Heterogeneous Bimodal Attention Fusion (HBAF) for multi-level multi-modal interaction in conversational emotion recognition. The proposed method comprises three key modules: the uni-modal representation module, the multi-modal fusion module, and the inter-modal contrastive learning module. The uni-modal representation module incorporates contextual content into low-level audio representations to bridge the heterogeneous multi-modal gap, enabling more effective fusion. The multi-modal fusion module uses dynamic bimodal attention and a dynamic gating mechanism to filter incorrect cross-modal relationships and fully exploit both intra-modal and inter-modal interactions. Finally, the inter-modal contrastive learning module captures complex absolute and relative interactions between audio and text modalities. Experiments on the MELD and IEMOCAP datasets demonstrate that the proposed HBAF method outperforms existing state-of-the-art baselines.