Small object detection (SOD) is a critical yet challenging task in computer vision, with applications like spanning surveillance, autonomous systems, medical imaging, and remote sensing. Unlike larger objects, small objects contain limited spatial and contextual information, making accurate detection difficult. Challenges such as low resolution, occlusion, background interference, and class imbalance further complicate the problem. This survey provides a comprehensive review of recent advancements in SOD using deep learning, focusing on articles published in Q1 journals during 2024-2025. We analyzed challenges, state-of-the-art techniques, datasets, evaluation metrics, and real-world applications. Recent advancements in deep learning have introduced innovative solutions, including multi-scale feature extraction, Super-Resolution (SR) techniques, attention mechanisms, and transformer-based architectures. Additionally, improvements in data augmentation, synthetic data generation, and transfer learning have addressed data scarcity and domain adaptation issues. Furthermore, emerging trends such as lightweight neural networks, knowledge distillation (KD), and self-supervised learning offer promising directions for improving detection efficiency, particularly in resource-constrained environments like Unmanned Aerial Vehicles (UAV)-based surveillance and edge computing. We also review widely used datasets, along with standard evaluation metrics such as mean Average Precision (mAP) and size-specific AP scores. The survey highlights real-world applications, including traffic monitoring, maritime surveillance, industrial defect detection, and precision agriculture. Finally, we discuss open research challenges and future directions, emphasizing the need for robust domain adaptation techniques, better feature fusion strategies, and real-time performance optimization.
Estimating motion in videos is an essential computer vision problem with many downstream applications, including controllable video generation and robotics. Current solutions are primarily trained using synthetic data or require tuning of situation-specific heuristics, which inherently limits these models' capabilities in real-world contexts. Despite recent developments in large-scale self-supervised learning from videos, leveraging such representations for motion estimation remains relatively underexplored. In this work, we develop Opt-CWM, a self-supervised technique for flow and occlusion estimation from a pre-trained next-frame prediction model. Opt-CWM works by learning to optimize counterfactual probes that extract motion information from a base video model, avoiding the need for fixed heuristics while training on unrestricted video inputs. We achieve state-of-the-art performance for motion estimation on real-world videos while requiring no labeled data.
Most existing contrastive learning-based sequential recommendation (SR) methods rely on random operations (e.g., crop, reorder, and substitute) to generate augmented sequences. These methods often struggle to create positive sample pairs that closely resemble the representations of the raw sequences, potentially disrupting item correlations by deleting key items or introducing noisy iterac, which misguides the contrastive learning process. To address this limitation, we propose Learnable sequence Augmentor for triplet Contrastive Learning in sequential Recommendation (LACLRec). Specifically, the self-supervised learning-based augmenter can automatically delete noisy items from sequences and insert new items that better capture item transition patterns, generating a higher-quality augmented sequence. Subsequently, we randomly generate another augmented sequence and design a ranking-based triplet contrastive loss to differentiate the similarities between the raw sequence, the augmented sequence from augmenter, and the randomly augmented sequence, providing more fine-grained contrastive signals. Extensive experiments on three real-world datasets demonstrate that both the sequence augmenter and the triplet contrast contribute to improving recommendation accuracy. LACLRec significantly outperforms the baseline model CL4SRec, and demonstrates superior performance compared to several state-of-the-art sequential recommendation algorithms.
The rise of deep learning has revolutionized data processing and prediction in signal processing and machine learning, yet the substantial computational demands of training and deploying modern large-scale deep models present significant challenges, including high computational costs and energy consumption. Recent research has uncovered a widespread phenomenon in deep networks: the emergence of low-rank structures in weight matrices and learned representations during training. These implicit low-dimensional patterns provide valuable insights for improving the efficiency of training and fine-tuning large-scale models. Practical techniques inspired by this phenomenon, such as low-rank adaptation (LoRA) and training, enable significant reductions in computational cost while preserving model performance. In this paper, we present a comprehensive review of recent advances in exploiting low-rank structures for deep learning and shed light on their mathematical foundations. Mathematically, we present two complementary perspectives on understanding the low-rankness in deep networks: (i) the emergence of low-rank structures throughout the whole optimization dynamics of gradient and (ii) the implicit regularization effects that induce such low-rank structures at convergence. From a practical standpoint, studying the low-rank learning dynamics of gradient descent offers a mathematical foundation for understanding the effectiveness of LoRA in fine-tuning large-scale models and inspires parameter-efficient low-rank training strategies. Furthermore, the implicit low-rank regularization effect helps explain the success of various masked training approaches in deep neural networks, ranging from dropout to masked self-supervised learning.
Whole-heart segmentation from CT and MRI scans is crucial for cardiovascular disease analysis, yet existing methods struggle with modality-specific biases and the need for extensive labeled datasets. To address these challenges, we propose a foundation model for whole-heart segmentation using a self-supervised learning (SSL) framework based on a student-teacher architecture. Our model is pretrained on a large, unlabeled dataset of CT and MRI scans, leveraging the xLSTM backbone to capture long-range spatial dependencies and complex anatomical structures in 3D medical images. By incorporating multi-modal pretraining, our approach ensures strong generalization across both CT and MRI modalities, mitigating modality-specific variations and improving segmentation accuracy in diverse clinical settings. The use of large-scale unlabeled data significantly reduces the dependency on manual annotations, enabling robust performance even with limited labeled data. We further introduce an xLSTM-UNet-based architecture for downstream whole-heart segmentation tasks, demonstrating its effectiveness on few-label CT and MRI datasets. Our results validate the robustness and adaptability of the proposed model, highlighting its potential for advancing automated whole-heart segmentation in medical imaging.
We propose Noisier2Inverse, a correction-free self-supervised deep learning approach for general inverse prob- lems. The proposed method learns a reconstruction function without the need for ground truth samples and is ap- plicable in cases where measurement noise is statistically correlated. This includes computed tomography, where detector imperfections or photon scattering create correlated noise patterns, as well as microscopy and seismic imaging, where physical interactions during measurement introduce dependencies in the noise structure. Similar to Noisier2Noise, a key step in our approach is the generation of noisier data from which the reconstruction net- work learns. However, unlike Noisier2Noise, the proposed loss function operates in measurement space and is trained to recover an extrapolated image instead of the original noisy one. This eliminates the need for an extrap- olation step during inference, which would otherwise suffer from ill-posedness. We numerically demonstrate that our method clearly outperforms previous self-supervised approaches that account for correlated noise.
In high-dimensional and high-stakes contexts, ensuring both rigorous statistical guarantees and interpretability in feature extraction from complex tabular data remains a formidable challenge. Traditional methods such as Principal Component Analysis (PCA) reduce dimensionality and identify key features that explain the most variance, but are constrained by their reliance on linear assumptions. In contrast, neural networks offer assumption-free feature extraction through self-supervised learning techniques such as autoencoders, though their interpretability remains a challenge in fields requiring transparency. To address this gap, this paper introduces Spofe, a novel self-supervised machine learning pipeline that marries the power of kernel principal components for capturing nonlinear dependencies with a sparse and principled polynomial representation to achieve clear interpretability with statistical rigor. Underpinning our approach is a robust theoretical framework that delivers precise error bounds and rigorous false discovery rate (FDR) control via a multi-objective knockoff selection procedure; it effectively bridges the gap between data-driven complexity and statistical reliability via three stages: (1) generating self-supervised signals using kernel principal components to model complex patterns, (2) distilling these signals into sparse polynomial functions for improved interpretability, and (3) applying a multi-objective knockoff selection procedure with significance testing to rigorously identify important features. Extensive experiments on diverse real-world datasets demonstrate the effectiveness of Spofe, consistently surpassing KPCA, SKPCA, and other methods in feature selection for regression and classification tasks. Visualization and case studies highlight its ability to uncover key insights, enhancing interpretability and practical utility.
The equivariant behaviour of features is essential in many computer vision tasks, yet popular self-supervised learning (SSL) methods tend to constrain equivariance by design. We propose a self-supervised learning approach where the system learns transformations independently by reconstructing images that have undergone previously unseen transformations. Specifically, the model is tasked to reconstruct intermediate transformed images, e.g. translated or rotated images, without prior knowledge of these transformations. This auxiliary task encourages the model to develop equivariance-coherent features without relying on predefined transformation rules. To this end, we apply transformations to the input image, generating an image pair, and then split the extracted features into two sets per image. One set is used with a usual SSL loss encouraging invariance, the other with our loss based on the auxiliary task to reconstruct the intermediate transformed images. Our loss and the SSL loss are linearly combined with weighted terms. Evaluating on synthetic tasks with natural images, our proposed method strongly outperforms all competitors, regardless of whether they are designed to learn equivariance. Furthermore, when trained alongside augmentation-based methods as the invariance tasks, such as iBOT or DINOv2, we successfully learn a balanced combination of invariant and equivariant features. Our approach performs strong on a rich set of realistic computer vision downstream tasks, almost always improving over all baselines.
Text images are unique in their dual nature, encompassing both visual and linguistic information. The visual component encompasses structural and appearance-based features, while the linguistic dimension incorporates contextual and semantic elements. In scenarios with degraded visual quality, linguistic patterns serve as crucial supplements for comprehension, highlighting the necessity of integrating both aspects for robust scene text recognition (STR). Contemporary STR approaches often use language models or semantic reasoning modules to capture linguistic features, typically requiring large-scale annotated datasets. Self-supervised learning, which lacks annotations, presents challenges in disentangling linguistic features related to the global context. Typically, sequence contrastive learning emphasizes the alignment of local features, while masked image modeling (MIM) tends to exploit local structures to reconstruct visual patterns, resulting in limited linguistic knowledge. In this paper, we propose a Linguistics-aware Masked Image Modeling (LMIM) approach, which channels the linguistic information into the decoding process of MIM through a separate branch. Specifically, we design a linguistics alignment module to extract vision-independent features as linguistic guidance using inputs with different visual appearances. As features extend beyond mere visual structures, LMIM must consider the global context to achieve reconstruction. Extensive experiments on various benchmarks quantitatively demonstrate our state-of-the-art performance, and attention visualizations qualitatively show the simultaneous capture of both visual and linguistic information.
Recent advances in self-supervised learning have led to the development of foundation models that have significantly advanced performance in various computer vision tasks. However, despite their potential, these models often overlook the crucial role of high-resolution digital surface models (DSMs) in understanding urban environments, particularly for building-level analysis, which is essential for applications like digital twins. To address this gap, we introduce HiRes-FusedMIM, a novel pre-trained model specifically designed to leverage the rich information contained within high-resolution RGB and DSM data. HiRes-FusedMIM utilizes a dual-encoder simple masked image modeling (SimMIM) architecture with a multi-objective loss function that combines reconstruction and contrastive objectives, enabling it to learn powerful, joint representations from both modalities. We conducted a comprehensive evaluation of HiRes-FusedMIM on a diverse set of downstream tasks, including classification, semantic segmentation, and instance segmentation. Our results demonstrate that: 1) HiRes-FusedMIM outperforms previous state-of-the-art geospatial methods on several building-related datasets, including WHU Aerial and LoveDA, demonstrating its effectiveness in capturing and leveraging fine-grained building information; 2) Incorporating DSMs during pre-training consistently improves performance compared to using RGB data alone, highlighting the value of elevation information for building-level analysis; 3) The dual-encoder architecture of HiRes-FusedMIM, with separate encoders for RGB and DSM data, significantly outperforms a single-encoder model on the Vaihingen segmentation task, indicating the benefits of learning specialized representations for each modality. To facilitate further research and applications in this direction, we will publicly release the trained model weights.
Despite advanced token compression techniques, existing multimodal large language models (MLLMs) still struggle with hour-long video understanding. In this work, we propose Video-XL-Pro, an efficient method for extremely long video understanding, built upon Reconstructive Compression of Tokens (ReCoT), a learnable module that leverages self-supervised learning to generate comprehensive and compact video tokens. ReCoT introduces two key components: (i) Dynamic Token Synthesizer (DTS): DTS generates pseudo-video tokens from static image tokens by learning intra-token relationships, which are then used in masked video modeling. (ii) Semantic-Guided Masking (SGM): SGM adaptively masks redundant visual tokens to facilitate more effective reconstructive learning. To improve training efficiency in MLLMs fine-tuning, we introduce a video-specific dataset pruning strategy and design a simple yet Query-aware Selector that enables the model to precisely locate query-relevant video tokens. With only 3B parameters, Video-XL-Pro outperforms most 7B models trained on larger datasets across multiple long video understanding benchmarks. Moreover, it can process over 8K frames on a single A100 GPU while maintaining high-quality performance.
According to the forgetting curve theory, we can enhance memory retention by learning extensive data and taking adequate rest. This means that in order to effectively retain new knowledge, it is essential to learn it thoroughly and ensure sufficient rest so that our brain can memorize without forgetting. The main takeaway from this theory is that learning extensive data at once necessitates sufficient rest before learning the same data again. This aspect of human long-term memory retention can be effectively utilized to address the continual learning of neural networks. Retaining new knowledge for a long period of time without catastrophic forgetting is the critical problem of continual learning. Therefore, based on Ebbinghaus' theory, we introduce the view-batch model that adjusts the learning schedules to optimize the recall interval between retraining the same samples. The proposed view-batch model allows the network to get enough rest to learn extensive knowledge from the same samples with a recall interval of sufficient length. To this end, we specifically present two approaches: 1) a replay method that guarantees the optimal recall interval, and 2) a self-supervised learning that acquires extensive knowledge from a single training sample at a time. We empirically show that these approaches of our method are aligned with the forgetting curve theory, which can enhance long-term memory. In our experiments, we also demonstrate that our method significantly improves many state-of-the-art continual learning methods in various protocols and scenarios. We open-source this project at https://github.com/hankyul2/ViewBatchModel.
Recognizing arbitrary or previously unseen categories is essential for comprehensive real-world 3D scene understanding. Currently, all existing methods rely on 2D or textual modalities during training, or together at inference. This highlights a clear absence of a model capable of processing 3D data alone for learning semantics end-to-end, along with the necessary data to train such a model. Meanwhile, 3D Gaussian Splatting (3DGS) has emerged as the de facto standard for 3D scene representation across various vision tasks. However, effectively integrating semantic reasoning into 3DGS in a generalizable fashion remains an open challenge. To address these limitations we introduce SceneSplat, to our knowledge the first large-scale 3D indoor scene understanding approach that operates natively on 3DGS. Furthermore, we propose a self-supervised learning scheme that unlocks rich 3D feature learning from unlabeled scenes. In order to power the proposed methods, we introduce SceneSplat-7K, the first large-scale 3DGS dataset for indoor scenes, comprising of 6868 scenes derived from 7 established datasets like ScanNet, Matterport3D, etc. Generating SceneSplat-7K required computational resources equivalent to 119 GPU-days on an L4 GPU, enabling standardized benchmarking for 3DGS-based reasoning for indoor scenes. Our exhaustive experiments on SceneSplat-7K demonstrate the significant benefit of the proposed methods over the established baselines.
Human activity recognition (HAR) with deep learning models relies on large amounts of labeled data, often challenging to obtain due to associated cost, time, and labor. Self-supervised learning (SSL) has emerged as an effective approach to leverage unlabeled data through pretext tasks, such as masked reconstruction and multitask learning with signal processing-based data augmentations, to pre-train encoder models. However, such methods are often derived from computer vision approaches that disregard physical mechanisms and constraints that govern wearable sensor data and the phenomena they reflect. In this paper, we propose a physics-informed multi-task pre-training (PIM) framework for IMU-based HAR. PIM generates pre-text tasks based on the understanding of basic physical aspects of human motion: including movement speed, angles of movement, and symmetry between sensor placements. Given a sensor signal, we calculate corresponding features using physics-based equations and use them as pretext tasks for SSL. This enables the model to capture fundamental physical characteristics of human activities, which is especially relevant for multi-sensor systems. Experimental evaluations on four HAR benchmark datasets demonstrate that the proposed method outperforms existing state-of-the-art methods, including data augmentation and masked reconstruction, in terms of accuracy and F1 score. We have observed gains of almost 10\% in macro f1 score and accuracy with only 2 to 8 labeled examples per class and up to 3% when there is no reduction in the amount of training data.
Contrastive learning in self-supervised settings primarily focuses on pre-training encoders, while decoders are typically introduced and trained separately for downstream dense prediction tasks. This conventional approach, however, overlooks the potential benefits of jointly pre-training both the encoder and decoder. In this paper, we propose DeCon: a framework-agnostic adaptation to convert an encoder-only self-supervised learning (SSL) contrastive approach to an efficient encoder-decoder framework that can be pre-trained in a contrastive manner. We first update the existing architecture to accommodate a decoder and its respective contrastive loss. We then introduce a weighted encoder-decoder contrastive loss with non-competing objectives that facilitates the joint encoder-decoder architecture pre-training. We adapt two established contrastive SSL frameworks tailored for dense prediction tasks, achieve new state-of-the-art results in COCO object detection and instance segmentation, and match state-of-the-art performance on Pascal VOC semantic segmentation. We show that our approach allows for pre-training a decoder and enhances the representation power of the encoder and its performance in dense prediction tasks. This benefit holds across heterogeneous decoder architectures between pre-training and fine-tuning and persists in out-of-domain, limited-data scenarios.
Remote Sensing Image Dehazing (RSID) poses significant challenges in real-world scenarios due to the complex atmospheric conditions and severe color distortions that degrade image quality. The scarcity of real-world remote sensing hazy image pairs has compelled existing methods to rely primarily on synthetic datasets. However, these methods struggle with real-world applications due to the inherent domain gap between synthetic and real data. To address this, we introduce Real-World Remote Sensing Hazy Image Dataset (RRSHID), the first large-scale dataset featuring real-world hazy and dehazed image pairs across diverse atmospheric conditions. Based on this, we propose MCAF-Net, a novel framework tailored for real-world RSID. Its effectiveness arises from three innovative components: Multi-branch Feature Integration Block Aggregator (MFIBA), which enables robust feature extraction through cascaded integration blocks and parallel multi-branch processing; Color-Calibrated Self-Supervised Attention Module (CSAM), which mitigates complex color distortions via self-supervised learning and attention-guided refinement; and Multi-Scale Feature Adaptive Fusion Module (MFAFM), which integrates features effectively while preserving local details and global context. Extensive experiments validate that MCAF-Net demonstrates state-of-the-art performance in real-world RSID, while maintaining competitive performance on synthetic datasets. The introduction of RRSHID and MCAF-Net sets new benchmarks for real-world RSID research, advancing practical solutions for this complex task. The code and dataset are publicly available at \url{https://github.com/lwCVer/RRSHID}.
Prediction tasks in digital pathology are challenging due to the massive size of whole-slide images (WSIs) and the weak nature of training signals. Advances in computing, data availability, and self-supervised learning (SSL) have paved the way for slide-level foundation models (SLFMs) that can improve prediction tasks in low-data regimes. However, working with these models is challenging, with issues such as catastrophic forgetting during fine-tuning and under-utilization of shared information between tasks and modalities. To overcome these two challenges, we propose ModalTune, a novel fine-tuning framework which introduces the Modal Adapter to integrate new modalities without modifying SLFM weights. Additionally, we use large-language models (LLMs) to encode labels as text, capturing semantic relationships and enhancing generalization across multiple tasks and cancer types in a single training recipe. ModalTune achieves state-of-the-art (SOTA) results against both uni-modal and multi-modal models across four cancer types, jointly improving survival and cancer subtype prediction while remaining competitive in pan-cancer settings. Additionally, we show ModalTune is highly generalizable to two out-of-distribution (OOD) datasets. To our knowledge, this is the first unified fine-tuning framework for multi-modal, multi-task, and pan-cancer modeling in digital pathology.
Deepfakes have become a universal and rapidly intensifying concern of generative AI across various media types such as images, audio, and videos. Among these, audio deepfakes have been of particular concern due to the ease of high-quality voice synthesis and distribution via platforms such as social media and robocalls. Consequently, detecting audio deepfakes plays a critical role in combating the growing misuse of AI-synthesized speech. However, real-world scenarios often introduce various audio corruptions, such as noise, modification, and compression, that may significantly impact detection performance. This work systematically evaluates the robustness of 10 audio deepfake detection models against 16 common corruptions, categorized into noise perturbation, audio modification, and compression. Using both traditional deep learning models and state-of-the-art foundation models, we make four unique observations. First, our findings show that while most models demonstrate strong robustness to noise, they are notably more vulnerable to modifications and compression, especially when neural codecs are applied. Second, speech foundation models generally outperform traditional models across most scenarios, likely due to their self-supervised learning paradigm and large-scale pre-training. Third, our results show that increasing model size improves robustness, albeit with diminishing returns. Fourth, we demonstrate how targeted data augmentation during training can enhance model resilience to unseen perturbations. A case study on political speech deepfakes highlights the effectiveness of foundation models in achieving high accuracy under real-world conditions. These findings emphasize the importance of developing more robust detection frameworks to ensure reliability in practical deployment settings.
Scaling up self-supervised learning has driven breakthroughs in language and vision, yet comparable progress has remained elusive in reinforcement learning (RL). In this paper, we study building blocks for self-supervised RL that unlock substantial improvements in scalability, with network depth serving as a critical factor. Whereas most RL papers in recent years have relied on shallow architectures (around 2 - 5 layers), we demonstrate that increasing the depth up to 1024 layers can significantly boost performance. Our experiments are conducted in an unsupervised goal-conditioned setting, where no demonstrations or rewards are provided, so an agent must explore (from scratch) and learn how to maximize the likelihood of reaching commanded goals. Evaluated on simulated locomotion and manipulation tasks, our approach increases performance by $2\times$ - $50\times$. Increasing the model depth not only increases success rates but also qualitatively changes the behaviors learned.
The past decade has witnessed notable achievements in self-supervised learning for video tasks. Recent efforts typically adopt the Masked Video Modeling (MVM) paradigm, leading to significant progress on multiple video tasks. However, two critical challenges remain: 1) Without human annotations, the random temporal sampling introduces uncertainty, increasing the difficulty of model training. 2) Previous MVM methods primarily recover the masked patches in the pixel space, leading to insufficient information compression for downstream tasks. To address these challenges jointly, we propose a self-supervised framework that leverages Temporal Correspondence for video Representation learning (T-CoRe). For challenge 1), we propose a sandwich sampling strategy that selects two auxiliary frames to reduce reconstruction uncertainty in a two-side-squeezing manner. Addressing challenge 2), we introduce an auxiliary branch into a self-distillation architecture to restore representations in the latent space, generating high-level semantic representations enriched with temporal information. Experiments of T-CoRe consistently present superior performance across several downstream tasks, demonstrating its effectiveness for video representation learning. The code is available at https://github.com/yafeng19/T-CORE.