Self-supervised learning with a contrastive batch approach has become a powerful tool for representation learning in computer vision. The performance of downstream tasks is proportional to the quality of visual features learned while self-supervised pre-training. The existing contrastive batch approaches heavily depend on data augmentation to learn latent information from unlabelled datasets. We argue that introducing the dataset’s intra-class variation in a contrastive batch approach improves visual representation quality further. In this paper, we propose a novel self-supervised learning approach named Iterative Pseudo-supervised Contrastive Learning (IPCL), which utilizes a balanced combination of image augmentations and pseudo-class information to improve the visual representation iteratively. Experimental results illustrate that our proposed method surpasses the baseline self-supervised method with the batch contrastive approach. It improves the visual representation quality over multiple datasets, leading to better performance on the downstream unsupervised image classification task.
The tactile sensation of textiles is critical in determining the comfort of clothing. For remote use, such as online shopping, users cannot physically touch the textile of clothes, making it difficult to evaluate its tactile sensation. Tactile sensing and actuation devices are required to transmit the tactile sensation of textiles. The sensing device needs to recognize different garments, even with hand-held sensors. In addition, the existing actuation device can only present a limited number of known patterns and cannot transmit unknown tactile sensations of textiles. To address these issues, we propose Telextiles, an interface that can remotely transmit tactile sensations of textiles by creating a latent space that reflects the proximity of textiles through contrastive self-supervised learning. We confirm that textiles with similar tactile features are located close to each other in the latent space through a two-dimensional plot. We then compress the latent features for known textile samples into the 1D distance and apply the 16 textile samples to the rollers in the order of the distance. The roller is rotated to select the textile with the closest feature if an unknown textile is detected.
Artificial neural networks trained on large, expert-labelled datasets are considered state-of-the-art for a range of medical image recognition tasks. However, categorically labelled datasets are time-consuming to generate and constrain classification to a pre-defined, fixed set of classes. For neuroradiological applications in particular, this represents a barrier to clinical adoption. To address these challenges, we present a self-supervised text-vision framework that learns to detect clinically relevant abnormalities in brain MRI scans by directly leveraging the rich information contained in accompanying free-text neuroradiology reports. Our training approach consisted of two-steps. First, a dedicated neuroradiological language model - NeuroBERT - was trained to generate fixed-dimensional vector representations of neuroradiology reports (N = 50,523) via domain-specific self-supervised learning tasks. Next, convolutional neural networks (one per MRI sequence) learnt to map individual brain scans to their corresponding text vector representations by optimising a mean square error loss. Once trained, our text-vision framework can be used to detect abnormalities in unreported brain MRI examinations by scoring scans against suitable query sentences (e.g., 'there is an acute stroke', 'there is hydrocephalus' etc.), enabling a range of classification-based applications including automated triage. Potentially, our framework could also serve as a clinical decision support tool, not only by suggesting findings to radiologists and detecting errors in provisional reports, but also by retrieving and displaying examples of pathologies from historical examinations that could be relevant to the current case based on textual descriptors.
Deep neural networks have reached remarkable achievements in medical image processing tasks, specifically in classifying and detecting various diseases. However, when confronted with limited data, these networks face a critical vulnerability, often succumbing to overfitting by excessively memorizing the limited information available. This work addresses the challenge mentioned above by improving the supervised contrastive learning method leveraging both image-level labels and domain-specific augmentations to enhance model robustness. This approach integrates self-supervised pre-training with a two-stage supervised contrastive learning strategy. In the first stage, we employ a modified supervised contrastive loss that not only focuses on reducing false negatives but also introduces an elimination effect to address false positives. In the second stage, a relaxing mechanism is introduced that refines positive and negative pairs based on similarity, ensuring that only relevant image representations are aligned. We evaluate our method on the BreakHis dataset, which consists of breast cancer histopathology images, and demonstrate an increase in classification accuracy by 1.45% in the image level, compared to the state-of-the-art method. This improvement corresponds to 93.63% absolute accuracy, highlighting the effectiveness of our approach in leveraging properties of data to learn more appropriate representation space.
The increasing proliferation of video surveillance cameras and the escalating demand for crime prevention have intensified interest in the task of violence detection within the research community. Compared to other action recognition tasks, violence detection in surveillance videos presents additional issues, such as the wide variety of real fight scenes. Unfortunately, existing datasets for violence detection are relatively small in comparison to those for other action recognition tasks. Moreover, surveillance footage often features different individuals in each video and varying backgrounds for each camera. In addition, fast detection of violent actions in real-life surveillance videos is crucial to prevent adverse outcomes, thus necessitating models that are optimized for reduced memory usage and computational costs. These challenges complicate the application of traditional action recognition methods. To tackle all these issues, we introduce JOSENet, a novel self-supervised framework that provides outstanding performance for violence detection in surveillance videos. The proposed model processes two spatiotemporal video streams, namely RGB frames and optical flows, and incorporates a new regularized self-supervised learning approach for videos. JOSENet demonstrates improved performance compared to state-of-the-art methods, while utilizing only one-fourth of the frames per video segment and operating at a reduced frame rate. The source code is available at https://github.com/ispamm/JOSENet.
We propose a hybrid framework for consistently producing high-quality object tracks by combining an automated object tracker with little human input. The key idea is to tailor a module for each dataset to intelligently decide when an object tracker is failing and so humans should be brought in to re-localize an object for continued tracking. Our approach leverages self-supervised learning on unlabeled videos to learn a tailored representation for a target object that is then used to actively monitor its tracked region and decide when the tracker fails. Since labeled data is not needed, our approach can be applied to novel object categories. Experiments on three datasets demonstrate our method outperforms existing approaches, especially for small, fast moving, or occluded objects.
The paper introduces AniTalker, an innovative framework designed to generate lifelike talking faces from a single portrait. Unlike existing models that primarily focus on verbal cues such as lip synchronization and fail to capture the complex dynamics of facial expressions and nonverbal cues, AniTalker employs a universal motion representation. This innovative representation effectively captures a wide range of facial dynamics, including subtle expressions and head movements. AniTalker enhances motion depiction through two self-supervised learning strategies: the first involves reconstructing target video frames from source frames within the same identity to learn subtle motion representations, and the second develops an identity encoder using metric learning while actively minimizing mutual information between the identity and motion encoders. This approach ensures that the motion representation is dynamic and devoid of identity-specific details, significantly reducing the need for labeled data. Additionally, the integration of a diffusion model with a variance adapter allows for the generation of diverse and controllable facial animations. This method not only demonstrates AniTalker's capability to create detailed and realistic facial movements but also underscores its potential in crafting dynamic avatars for real-world applications. Synthetic results can be viewed at https://github.com/X-LANCE/AniTalker.
In this paper, we consider two challenging issues in reference-based super-resolution (RefSR) for smartphone, (i) how to choose a proper reference image, and (ii) how to learn RefSR in a self-supervised manner. Particularly, we propose a novel self-supervised learning approach for real-world RefSR from observations at dual and multiple camera zooms. Firstly, considering the popularity of multiple cameras in modern smartphones, the more zoomed (telephoto) image can be naturally leveraged as the reference to guide the super-resolution (SR) of the lesser zoomed (ultra-wide) image, which gives us a chance to learn a deep network that performs SR from the dual zoomed observations (DZSR). Secondly, for self-supervised learning of DZSR, we take the telephoto image instead of an additional high-resolution image as the supervision information, and select a center patch from it as the reference to super-resolve the corresponding ultra-wide image patch. To mitigate the effect of the misalignment between ultra-wide low-resolution (LR) patch and telephoto ground-truth (GT) image during training, we first adopt patch-based optical flow alignment and then design an auxiliary-LR to guide the deforming of the warped LR features. To generate visually pleasing results, we present local overlapped sliced Wasserstein loss to better represent the perceptual difference between GT and output in the feature space. During testing, DZSR can be directly deployed to super-solve the whole ultra-wide image with the reference of the telephoto image. In addition, we further take multiple zoomed observations to explore self-supervised RefSR, and present a progressive fusion scheme for the effective utilization of reference images. Experiments show that our methods achieve better quantitative and qualitative performance against state-of-the-arts. Codes are available at https://github.com/cszhilu1998/SelfDZSR_PlusPlus.
Self-supervised learning (SSL) has emerged as a key technique for training networks that can generalize well to diverse tasks without task-specific supervision. This property makes SSL desirable for computational pathology, the study of digitized images of tissues, as there are many target applications and often limited labeled training samples. However, SSL algorithms and models have been primarily developed in the field of natural images and whether their performance can be improved by adaptation to particular domains remains an open question. In this work, we present an investigation of modifications to SSL for pathology data, specifically focusing on the DINOv2 algorithm. We propose alternative augmentations, regularization functions, and position encodings motivated by the characteristics of pathology images. We evaluate the impact of these changes on several benchmarks to demonstrate the value of tailored approaches.
In this paper, we present a novel approach for text independent phone-to-audio alignment based on phoneme recognition, representation learning and knowledge transfer. Our method leverages a self-supervised model (wav2vec2) fine-tuned for phoneme recognition using a Connectionist Temporal Classification (CTC) loss, a dimension reduction model and a frame-level phoneme classifier trained thanks to forced-alignment labels (using Montreal Forced Aligner) to produce multi-lingual phonetic representations, thus requiring minimal additional training. We evaluate our model using synthetic native data from the TIMIT dataset and the SCRIBE dataset for American and British English, respectively. Our proposed model outperforms the state-of-the-art (charsiu) in statistical metrics and has applications in language learning and speech processing systems. We leave experiments on other languages for future work but the design of the system makes it easily adaptable to other languages.
An accurate detection and tracking of devices such as guiding catheters in live X-ray image acquisitions is an essential prerequisite for endovascular cardiac interventions. This information is leveraged for procedural guidance, e.g., directing stent placements. To ensure procedural safety and efficacy, there is a need for high robustness no failures during tracking. To achieve that, one needs to efficiently tackle challenges, such as: device obscuration by contrast agent or other external devices or wires, changes in field-of-view or acquisition angle, as well as the continuous movement due to cardiac and respiratory motion. To overcome the aforementioned challenges, we propose a novel approach to learn spatio-temporal features from a very large data cohort of over 16 million interventional X-ray frames using self-supervision for image sequence data. Our approach is based on a masked image modeling technique that leverages frame interpolation based reconstruction to learn fine inter-frame temporal correspondences. The features encoded in the resulting model are fine-tuned downstream. Our approach achieves state-of-the-art performance and in particular robustness compared to ultra optimized reference solutions (that use multi-stage feature fusion, multi-task and flow regularization). The experiments show that our method achieves 66.31% reduction in maximum tracking error against reference solutions (23.20% when flow regularization is used); achieving a success score of 97.95% at a 3x faster inference speed of 42 frames-per-second (on GPU). The results encourage the use of our approach in various other tasks within interventional image analytics that require effective understanding of spatio-temporal semantics.
Self-supervised learning for image denoising problems in the presence of denaturation for noisy data is a crucial approach in machine learning. However, theoretical understanding of the performance of the approach that uses denatured data is lacking. To provide better understanding of the approach, in this paper, we analyze a self-supervised denoising algorithm that uses denatured data in depth through theoretical analysis and numerical experiments. Through the theoretical analysis, we discuss that the algorithm finds desired solutions to the optimization problem with the population risk, while the guarantee for the empirical risk depends on the hardness of the denoising task in terms of denaturation levels. We also conduct several experiments to investigate the performance of an extended algorithm in practice. The results indicate that the algorithm training with denatured images works, and the empirical performance aligns with the theoretical results. These results suggest several insights for further improvement of self-supervised image denoising that uses denatured data in future directions.
In this paper, we investigate what constitutes a good representation or model in self-supervised learning (SSL). We argue that a good representation should exhibit universality, characterized by three essential properties: discriminability, generalizability, and transferability. While these capabilities are implicitly desired in most SSL frameworks, existing methods lack an explicit modeling of universality, and its theoretical foundations remain underexplored. To address these gaps, we propose General SSL (GeSSL), a novel framework that explicitly models universality from three complementary dimensions: the optimization objective, the parameter update mechanism, and the learning paradigm. GeSSL integrates a bi-level optimization structure that jointly models task-specific adaptation and cross-task consistency, thereby capturing all three aspects of universality within a unified SSL objective. Furthermore, we derive a theoretical generalization bound, ensuring that the optimization process of GeSSL consistently leads to representations that generalize well to unseen tasks. Empirical results on multiple benchmark datasets demonstrate that GeSSL consistently achieves superior performance across diverse downstream tasks, validating its effectiveness in modeling universal representations.
Background and Purpose: Identifying the thromboembolism source in ischemic stroke is crucial for treatment and secondary prevention yet is often undetermined. This study describes a self-supervised deep learning approach in digital pathology of emboli for classifying ischemic stroke clot origin from histopathological images. Methods: The dataset included whole slide images (WSI) from the STRIP AI Kaggle challenge, consisting of retrieved clots from ischemic stroke patients following mechanical thrombectomy. Transformer-based deep learning models were developed using transfer learning and self-supervised pretraining for classifying WSI. Customizations included an attention pooling layer, weighted loss function, and threshold optimization. Various model architectures were tested and compared, and model performances were primarily evaluated using weighted logarithmic loss. Results: The model achieved a logloss score of 0.662 in cross-validation and 0.659 on the test set. Different model backbones were compared, with the swin_large_patch4_window12_384 showed higher performance. Thresholding techniques for clot origin classification were employed to balance false positives and negatives. Conclusion: The study demonstrates the extent of efficacy of transformer-based deep learning models in identifying ischemic stroke clot origins from histopathological images and emphasizes the need for refined modeling techniques specifically adapted to thrombi WSI. Further research is needed to improve model performance, interpretability, validate its effectiveness. Future enhancement could include integrating larger patient cohorts, advanced preprocessing strategies, and exploring ensemble multimodal methods for enhanced diagnostic accuracy.
AI Foundation models are gaining traction in various applications, including medical fields like radiology. However, medical foundation models are often tested on limited tasks, leaving their generalisability and biases unexplored. We present RayDINO, a large visual encoder trained by self-supervision on 873k chest X-rays. We compare RayDINO to previous state-of-the-art models across nine radiology tasks, from classification and dense segmentation to text generation, and provide an in depth analysis of population, age and sex biases of our model. Our findings suggest that self-supervision allows patient-centric AI proving useful in clinical workflows and interpreting X-rays holistically. With RayDINO and small task-specific adapters, we reach state-of-the-art results and improve generalization to unseen populations while mitigating bias, illustrating the true promise of foundation models: versatility and robustness.
A wide range of graph embedding objectives decompose into two components: one that enforces similarity, attracting the embeddings of nodes that are perceived as similar, and another that enforces dissimilarity, repelling the embeddings of nodes that are perceived as dissimilar. Without repulsion, the embeddings would collapse into trivial solutions. Skip-Gram Negative Sampling (SGNS) is a popular and efficient repulsion approach that prevents collapse by repelling each node from a sample of dissimilar nodes. In this work, we show that when repulsion is most needed and the embeddings approach collapse, SGNS node-wise repulsion is, in the aggregate, an approximate re-centering of the node embedding dimensions. Such dimension operations are more scalable than node operations and produce a simpler geometric interpretation of the repulsion. Our theoretical result establishes dimension regularization as an effective and more efficient, compared to skip-gram node contrast, approach to enforcing dissimilarity among embeddings of nodes. We use this result to propose a flexible algorithm augmentation framework that improves the scalability of any existing algorithm using SGNS. The framework prioritizes node attraction and replaces SGNS with dimension regularization. We instantiate this generic framework for LINE and node2vec and show that the augmented algorithms preserve downstream link-prediction performance while reducing GPU memory usage by up to 33.3% and training time by 23.4%. Moreover, we show that completely removing repulsion (a special case of our augmentation framework) in LINE reduces training time by 70.9% on average, while increasing link prediction performance, especially for graphs that are globally sparse but locally dense. In general, however, repulsion is needed, and dimension regularization provides an efficient alternative to SGNS.
This paper investigates the effectiveness of self-supervised pre-trained vision transformers (ViTs) compared to supervised pre-trained ViTs and conventional neural networks (ConvNets) for detecting facial deepfake images and videos. It examines their potential for improved generalization and explainability, especially with limited training data. Despite the success of transformer architectures in various tasks, the deepfake detection community is hesitant to use large ViTs as feature extractors due to their perceived need for extensive data and suboptimal generalization with small datasets. This contrasts with ConvNets, which are already established as robust feature extractors. Additionally, training ViTs from scratch requires significant resources, limiting their use to large companies. Recent advancements in self-supervised learning (SSL) for ViTs, like masked autoencoders and DINOs, show adaptability across diverse tasks and semantic segmentation capabilities. By leveraging SSL ViTs for deepfake detection with modest data and partial fine-tuning, we find comparable adaptability to deepfake detection and explainability via the attention mechanism. Moreover, partial fine-tuning of ViTs is a resource-efficient option.
Deep neural networks, including Convolutional Neural Networks (CNNs) and Visual Transformers (ViT), have achieved stunning success in medical image domain. We study thorax disease classification in this paper. Effective extraction of features for the disease areas is crucial for disease classification on radiographic images. While various neural architectures and training techniques, such as self-supervised learning with contrastive/restorative learning, have been employed for disease classification on radiographic images, there are no principled methods which can effectively reduce the adverse effect of noise and background, or non-disease areas, on the radiographic images for disease classification. To address this challenge, we propose a novel Low-Rank Feature Learning (LRFL) method in this paper, which is universally applicable to the training of all neural networks. The LRFL method is both empirically motivated by the low frequency property observed on all the medical datasets in this paper, and theoretically motivated by our sharp generalization bound for neural networks with low-rank features. In the empirical study, using a neural network such as a ViT or a CNN pre-trained on unlabeled chest X-rays by Masked Autoencoders (MAE), our novel LRFL method is applied on the pre-trained neural network and demonstrate better classification results in terms of both multiclass area under the receiver operating curve (mAUC) and classification accuracy.
Foundation models, especially those using transformers as backbones, have gained significant popularity, particularly in language and language-vision tasks. However, large foundation models are typically trained on high-quality data, which poses a significant challenge, given the prevalence of poor-quality real-world data. This challenge is more pronounced for developing foundation models for physiological data; such data are often noisy, incomplete, or inconsistent. The present work aims to provide a toolset for developing foundation models on physiological data. We leverage a large dataset of photoplethysmography (PPG) signals from hospitalized intensive care patients. For this data, we propose SimQuality, a novel self-supervised learning task based on convolutional neural networks (CNNs) as the backbone to enforce representations to be similar for good and poor quality signals that are from similar physiological states. We pre-trained the SimQuality on over 36 million 30-second PPG pairs and then fine-tuned and tested on six downstream tasks using external datasets. The results demonstrate the superiority of the proposed approach on all the downstream tasks, which are extremely important for heart monitoring on wearable devices. Our method indicates that CNNs can be an effective backbone for foundation models that are robust to training data quality.
Self-supervised learning (SSL) has developed rapidly in recent years. However, most of the mainstream methods are computationally expensive and rely on two (or more) augmentations for each image to construct positive pairs. Moreover, they mainly focus on large models and large-scale datasets, which lack flexibility and feasibility in many practical applications. In this paper, we propose an efficient single-branch SSL method based on non-parametric instance discrimination, aiming to improve the algorithm, model, and data efficiency of SSL. By analyzing the gradient formula, we correct the update rule of the memory bank with improved performance. We further propose a novel self-distillation loss that minimizes the KL divergence between the probability distribution and its square root version. We show that this alleviates the infrequent updating problem in instance discrimination and greatly accelerates convergence. We systematically compare the training overhead and performance of different methods in different scales of data, and under different backbones. Experimental results show that our method outperforms various baselines with significantly less overhead, and is especially effective for limited amounts of data and small models.