Multi-modal ophthalmic image classification plays a key role in diagnosing eye diseases, as it integrates information from different sources to complement their respective performances. However, recent improvements have mainly focused on accuracy, often neglecting the importance of confidence and robustness in predictions for diverse modalities. In this study, we propose a novel multi-modality evidential fusion pipeline for eye disease screening. It provides a measure of confidence for each modality and elegantly integrates the multi-modality information using a multi-distribution fusion perspective. Specifically, our method first utilizes normal inverse gamma prior distributions over pre-trained models to learn both aleatoric and epistemic uncertainty for uni-modality. Then, the normal inverse gamma distribution is analyzed as the Student's t distribution. Furthermore, within a confidence-aware fusion framework, we propose a mixture of Student's t distributions to effectively integrate different modalities, imparting the model with heavy-tailed properties and enhancing its robustness and reliability. More importantly, the confidence-aware multi-modality ranking regularization term induces the model to more reasonably rank the noisy single-modal and fused-modal confidence, leading to improved reliability and accuracy. Experimental results on both public and internal datasets demonstrate that our model excels in robustness, particularly in challenging scenarios involving Gaussian noise and modality missing conditions. Moreover, our model exhibits strong generalization capabilities to out-of-distribution data, underscoring its potential as a promising solution for multimodal eye disease screening.
Visual single object tracking aims to continuously localize and estimate the scale of a target in subsequent video frames, given only its initial state in the first frame. This task has traditionally been framed as a template matching problem, evolving through major phases including correlation filters, two-stream networks, and one-stream networks with significant progress achieved. However, these methods typically require explicit classification and regression modeling, depend on supervised training with large-scale datasets, and are limited to the single task of tracking, lacking flexibility. In recent years, multi-modal large language models (MLLMs) have advanced rapidly. Open-source models like Qwen2.5-VL, a flagship MLLMs with strong foundational capabilities, demonstrate excellent performance in grounding tasks. This has spurred interest in applying such models directly to visual tracking. However, experiments reveal that Qwen2.5-VL struggles with template matching between image pairs (i.e., tracking tasks). Inspired by deepseek-R1, we fine-tuned Qwen2.5-VL using the group relative policy optimization (GRPO) reinforcement learning method on a small-scale dataset with a rule-based reward function. The resulting model, R1-Track, achieved notable performance on the GOT-10k benchmark. R1-Track supports flexible initialization via bounding boxes or text descriptions while retaining most of the original model's general capabilities. And we further discuss potential improvements for R1-Track. This rough technical report summarizes our findings as of May 2025.
Exploiting the wealth of imaging and non-imaging information for disease prediction tasks requires models capable of representing, at the same time, individual features as well as data associations between subjects from potentially large populations. Graphs provide a natural framework for such tasks, yet previous graph-based approaches focus on pairwise similarities without modelling the subjects' individual characteristics and features. On the other hand, relying solely on subject-specific imaging feature vectors fails to model the interaction and similarity between subjects, which can reduce performance. In this paper, we introduce the novel concept of Graph Convolutional Networks (GCN) for brain analysis in populations, combining imaging and non-imaging data. We represent populations as a sparse graph where its vertices are associated with image-based feature vectors and the edges encode phenotypic information. This structure was used to train a GCN model on partially labelled graphs, aiming to infer the classes of unlabelled nodes from the node features and pairwise associations between subjects. We demonstrate the potential of the method on the challenging ADNI and ABIDE databases, as a proof of concept of the benefit from integrating contextual information in classification tasks. This has a clear impact on the quality of the predictions, leading to 69.5% accuracy for ABIDE (outperforming the current state of the art of 66.8%) and 77% for ADNI for prediction of MCI conversion, significantly outperforming standard linear classifiers where only individual features are considered.
Most signal processing problems involve the challenging task of multidimensional probability density function (PDF) estimation. In this work, we propose a solution to this problem by using a family of Rotation-based Iterative Gaussianization (RBIG) transforms. The general framework consists of the sequential application of a univariate marginal Gaussianization transform followed by an orthonormal transform. The proposed procedure looks for differentiable transforms to a known PDF so that the unknown PDF can be estimated at any point of the original domain. In particular, we aim at a zero mean unit covariance Gaussian for convenience. RBIG is formally similar to classical iterative Projection Pursuit (PP) algorithms. However, we show that, unlike in PP methods, the particular class of rotations used has no special qualitative relevance in this context, since looking for interestingness is not a critical issue for PDF estimation. The key difference is that our approach focuses on the univariate part (marginal Gaussianization) of the problem rather than on the multivariate part (rotation). This difference implies that one may select the most convenient rotation suited to each practical application. The differentiability, invertibility and convergence of RBIG are theoretically and experimentally analyzed. Relation to other methods, such as Radial Gaussianization (RG), one-class support vector domain description (SVDD), and deep neural networks (DNN) is also pointed out. The practical performance of RBIG is successfully illustrated in a number of multidimensional problems such as image synthesis, classification, denoising, and multi-information estimation.
Regional dropout strategies have been proposed to enhance the performance of convolutional neural network classifiers. They have proved to be effective for guiding the model to attend on less discriminative parts of objects (e.g. leg as opposed to head of a person), thereby letting the network generalize better and have better object localization capabilities. On the other hand, current methods for regional dropout remove informative pixels on training images by overlaying a patch of either black pixels or random noise. Such removal is not desirable because it leads to information loss and inefficiency during training. We therefore propose the CutMix augmentation strategy: patches are cut and pasted among training images where the ground truth labels are also mixed proportionally to the area of the patches. By making efficient use of training pixels and retaining the regularization effect of regional dropout, CutMix consistently outperforms the state-of-the-art augmentation strategies on CIFAR and ImageNet classification tasks, as well as on the ImageNet weakly-supervised localization task. Moreover, unlike previous augmentation methods, our CutMix-trained ImageNet classifier, when used as a pretrained model, results in consistent performance gains in Pascal detection and MS-COCO image captioning benchmarks. We also show that CutMix improves the model robustness against input corruptions and its out-of-distribution detection performances. Source code and pretrained models are available at \href{https://github.com/clovaai/CutMix-PyTorch}{https://github.com/clovaai/CutMix-PyTorch}.
This paper presents a comprehensive survey on vision-based robotic grasping. We conclude three key tasks during vision-based robotic grasping, which are object localization, object pose estimation and grasp estimation. In detail, the object localization task contains object localization without classification, object detection and object instance segmentation. This task provides the regions of the target object in the input data. The object pose estimation task mainly refers to estimating the 6D object pose and includes correspondence-based methods, template-based methods and voting-based methods, which affords the generation of grasp poses for known objects. The grasp estimation task includes 2D planar grasp methods and 6DoF grasp methods, where the former is constrained to grasp from one direction. These three tasks could accomplish the robotic grasping with different combinations. Lots of object pose estimation methods need not object localization, and they conduct object localization and object pose estimation jointly. Lots of grasp estimation methods need not object localization and object pose estimation, and they conduct grasp estimation in an end-to-end manner. Both traditional methods and latest deep learning-based methods based on the RGB-D image inputs are reviewed elaborately in this survey. Related datasets and comparisons between state-of-the-art methods are summarized as well. In addition, challenges about vision-based robotic grasping and future directions in addressing these challenges are also pointed out.
Domain-specific data is the crux of the successful transfer of machine learning systems from benchmarks to real life. In simple problems such as image classification, crowdsourcing has become one of the standard tools for cheap and time-efficient data collection: thanks in large part to advances in research on aggregation methods. However, the applicability of crowdsourcing to more complex tasks (e.g., speech recognition) remains limited due to the lack of principled aggregation methods for these modalities. The main obstacle towards designing aggregation methods for more advanced applications is the absence of training data, and in this work, we focus on bridging this gap in speech recognition. For this, we collect and release CrowdSpeech -- the first publicly available large-scale dataset of crowdsourced audio transcriptions. Evaluation of existing and novel aggregation methods on our data shows room for improvement, suggesting that our work may entail the design of better algorithms. At a higher level, we also contribute to the more general challenge of developing the methodology for reliable data collection via crowdsourcing. In that, we design a principled pipeline for constructing datasets of crowdsourced audio transcriptions in any novel domain. We show its applicability on an under-resourced language by constructing VoxDIY -- a counterpart of CrowdSpeech for the Russian language. We also release the code that allows a full replication of our data collection pipeline and share various insights on best practices of data collection via crowdsourcing.
Model poisoning attacks on federated learning (FL) intrude in the entire system via compromising an edge model, resulting in malfunctioning of machine learning models. Such compromised models are tampered with to perform adversary-desired behaviors. In particular, we considered a semi-targeted situation where the source class is predetermined however the target class is not. The goal is to cause the global classifier to misclassify data of the source class. Though approaches such as label flipping have been adopted to inject poisoned parameters into FL, it has been shown that their performances are usually class-sensitive varying with different target classes applied. Typically, an attack can become less effective when shifting to a different target class. To overcome this challenge, we propose the Attacking Distance-aware Attack (ADA) to enhance a poisoning attack by finding the optimized target class in the feature space. Moreover, we studied a more challenging situation where an adversary had limited prior knowledge about a client's data. To tackle this problem, ADA deduces pair-wise distances between different classes in the latent feature space from shared model parameters based on the backward error analysis. We performed extensive empirical evaluations on ADA by varying the factor of attacking frequency in three different image classification tasks. As a result, ADA succeeded in increasing the attack performance by 1.8 times in the most challenging case with an attacking frequency of 0.01.
Handwriting recognition is one of the active and challenging areas of research in the field of image processing and pattern recognition. It has many applications that include: a reading aid for visual impairment, automated reading and processing for bank checks, making any handwritten document searchable, and converting them into structural text form, etc. Moreover, high accuracy rates have been recorded by handwriting recognition systems for English, Chinese Arabic, Persian, and many other languages. Yet there is no such system available for offline Kurdish handwriting recognition. In this paper, an attempt is made to design and develop a model that can recognize handwritten characters for Kurdish alphabets using deep learning techniques. Kurdish (Sorani) contains 34 characters and mainly employs an Arabic\Persian based script with modified alphabets. In this work, a Deep Convolutional Neural Network model is employed that has shown exemplary performance in handwriting recognition systems. Then, a comprehensive dataset was created for handwritten Kurdish characters, which contains more than 40 thousand images. The created dataset has been used for training the Deep Convolutional Neural Network model for classification and recognition tasks. In the proposed system, the experimental results show an acceptable recognition level. The testing results reported a 96% accuracy rate, and training accuracy reported a 97% accuracy rate. From the experimental results, it is clear that the proposed deep learning model is performing well and is comparable to the similar model of other languages' handwriting recognition systems.
Semi-supervised learning frameworks usually adopt mutual learning approaches with multiple submodels to learn from different perspectives. To avoid transferring erroneous pseudo labels between these submodels, a high threshold is usually used to filter out a large number of low-confidence predictions for unlabeled data. However, such filtering can not fully exploit unlabeled data with low prediction confidence. To overcome this problem, in this work, we propose a mutual learning framework based on pseudo-negative labels. Negative labels are those that a corresponding data item does not belong. In each iteration, one submodel generates pseudo-negative labels for each data item, and the other submodel learns from these labels. The role of the two submodels exchanges after each iteration until convergence. By reducing the prediction probability on pseudo-negative labels, the dual model can improve its prediction ability. We also propose a mechanism to select a few pseudo-negative labels to feed into submodels. In the experiments, our framework achieves state-of-the-art results on several main benchmarks. Specifically, with our framework, the error rates of the 13-layer CNN model are 9.35% and 7.94% for CIFAR-10 with 1000 and 4000 labels, respectively. In addition, for the non-augmented MNIST with only 20 labels, the error rate is 0.81% by our framework, which is much smaller than that of other approaches. Our approach also demonstrates a significant performance improvement in domain adaptation.
As research interests in medical image analysis become increasingly fine-grained, the cost for extensive annotation also rises. One feasible way to reduce the cost is to annotate with coarse-grained superclass labels while using limited fine-grained annotations as a complement. In this way, fine-grained data learning is assisted by ample coarse annotations. Recent studies in classification tasks have adopted this method to achieve satisfactory results. However, there is a lack of research on efficient learning of fine-grained subclasses in semantic segmentation tasks. In this paper, we propose a novel approach that leverages the hierarchical structure of categories to design network architecture. Meanwhile, a task-driven data generation method is presented to make it easier for the network to recognize different subclass categories. Specifically, we introduce a Prior Concatenation module that enhances confidence in subclass segmentation by concatenating predicted logits from the superclass classifier, a Separate Normalization module that stretches the intra-class distance within the same superclass to facilitate subclass segmentation, and a HierarchicalMix model that generates high-quality pseudo labels for unlabeled samples by fusing only similar superclass regions from labeled and unlabeled images. Our experiments on the BraTS2021 and ACDC datasets demonstrate that our approach achieves comparable accuracy to a model trained with full subclass annotations, with limited subclass annotations and sufficient superclass annotations. Our approach offers a promising solution for efficient fine-grained subclass segmentation in medical images. Our code is publicly available here.
Vision foundation models such as Contrastive Vision-Language Pre-training (CLIP) and Segment Anything (SAM) have demonstrated impressive zero-shot performance on image classification and segmentation tasks. However, the incorporation of CLIP and SAM for label-free scene understanding has yet to be explored. In this paper, we investigate the potential of vision foundation models in enabling networks to comprehend 2D and 3D worlds without labelled data. The primary challenge lies in effectively supervising networks under extremely noisy pseudo labels, which are generated by CLIP and further exacerbated during the propagation from the 2D to the 3D domain. To tackle these challenges, we propose a novel Cross-modality Noisy Supervision (CNS) method that leverages the strengths of CLIP and SAM to supervise 2D and 3D networks simultaneously. In particular, we introduce a prediction consistency regularization to co-train 2D and 3D networks, then further impose the networks' latent space consistency using the SAM's robust feature representation. Experiments conducted on diverse indoor and outdoor datasets demonstrate the superior performance of our method in understanding 2D and 3D open environments. Our 2D and 3D network achieves label-free semantic segmentation with 28.4\% and 33.5\% mIoU on ScanNet, improving 4.7\% and 7.9\%, respectively. For nuImages and nuScenes datasets, the performance is 22.1\% and 26.8\% with improvements of 3.5\% and 6.0\%, respectively. Code is available. (https://github.com/runnanchen/Label-Free-Scene-Understanding).
Zero-shot Human-Object Interaction (HOI) detection has emerged as a frontier topic due to its capability to detect HOIs beyond a predefined set of categories. This task entails not only identifying the interactiveness of human-object pairs and localizing them but also recognizing both seen and unseen interaction categories. In this paper, we introduce a novel framework for zero-shot HOI detection using Conditional Multi-Modal Prompts, namely CMMP. This approach enhances the generalization of large foundation models, such as CLIP, when fine-tuned for HOI detection. Unlike traditional prompt-learning methods, we propose learning decoupled vision and language prompts for interactiveness-aware visual feature extraction and generalizable interaction classification, respectively. Specifically, we integrate prior knowledge of different granularity into conditional vision prompts, including an input-conditioned instance prior and a global spatial pattern prior. The former encourages the image encoder to treat instances belonging to seen or potentially unseen HOI concepts equally while the latter provides representative plausible spatial configuration of the human and object under interaction. Besides, we employ language-aware prompt learning with a consistency constraint to preserve the knowledge of the large foundation model to enable better generalization in the text branch. Extensive experiments demonstrate the efficacy of our detector with conditional multi-modal prompts, outperforming previous state-of-the-art on unseen classes of various zero-shot settings. The code and models are available at \url{https://github.com/ltttpku/CMMP}.
The right to be forgotten is a fundamental principle of privacy-preserving regulations and extends to Machine Learning (ML) paradigms such as Federated Learning (FL). While FL enhances privacy by enabling collaborative model training without sharing private data, trained models still retain the influence of training data. Federated Unlearning (FU) methods recently proposed often rely on impractical assumptions for real-world FL deployments, such as storing client update histories or requiring access to a publicly available dataset. To address these constraints, this paper introduces a novel method that leverages negated Pseudo-gradients Updates for Federated Unlearning (PUF). Our approach only uses standard client model updates, anyway employed during regular FL rounds, and interprets them as pseudo-gradients. When a client needs to be forgotten, we apply the negated of their pseudo-gradients, appropriately scaled, to the global model. Unlike state-of-the-art mechanisms, PUF seamlessly integrates with FL workflows, incurs no additional computational and communication overhead beyond standard FL rounds, and supports concurrent unlearning requests. We extensively evaluated the proposed method on two well-known benchmark image classification datasets (CIFAR-10 and CIFAR-100) and a real-world medical imaging dataset for segmentation (ProstateMRI), using three different neural architectures: two residual networks and a vision transformer. The experimental results across various settings demonstrate that PUF achieves state-of-the-art forgetting effectiveness and recovery time, without relying on any additional assumptions, thus underscoring its practical applicability.
The vulnerability of deep networks to adversarial attacks is a central problem for deep learning from the perspective of both cognition and security. The current most successful defense method is to train a classifier using adversarial images created during learning. Another defense approach involves transformation or purification of the original input to remove adversarial signals before the image is classified. We focus on defending naturally-trained classifiers using Markov Chain Monte Carlo (MCMC) sampling with an Energy-Based Model (EBM) for adversarial purification. In contrast to adversarial training, our approach is intended to secure pre-existing and highly vulnerable classifiers. The memoryless behavior of long-run MCMC sampling will eventually remove adversarial signals, while metastable behavior preserves consistent appearance of MCMC samples after many steps to allow accurate long-run prediction. Balancing these factors can lead to effective purification and robust classification. We evaluate adversarial defense with an EBM using the strongest known attacks against purification. Our contributions are 1) an improved method for training EBM's with realistic long-run MCMC samples, 2) an Expectation-Over-Transformation (EOT) defense that resolves theoretical ambiguities for stochastic defenses and from which the EOT attack naturally follows, and 3) state-of-the-art adversarial defense for naturally-trained classifiers and competitive defense compared to adversarially-trained classifiers on Cifar-10, SVHN, and Cifar-100. Code and pre-trained models are available at https://github.com/point0bar1/ebm-defense.
Refraining from confidently predicting when faced with categories of inputs different from those seen during training is an important requirement for the safe deployment of deep learning systems. While simple to state, this has been a particularly challenging problem in deep learning, where models often end up making overconfident predictions in such situations. In this work we present a simple, but highly effective approach to deal with out-of-distribution detection that uses the principle of abstention: when encountering a sample from an unseen class, the desired behavior is to abstain from predicting. Our approach uses a network with an extra abstention class and is trained on a dataset that is augmented with an uncurated set that consists of a large number of out-of-distribution (OoD) samples that are assigned the label of the abstention class; the model is then trained to learn an effective discriminator between in and out-of-distribution samples. We compare this relatively simple approach against a wide variety of more complex methods that have been proposed both for out-of-distribution detection as well as uncertainty modeling in deep learning, and empirically demonstrate its effectiveness on a wide variety of of benchmarks and deep architectures for image recognition and text classification, often outperforming existing approaches by significant margins. Given the simplicity and effectiveness of this method, we propose that this approach be used as a new additional baseline for future work in this domain.
With the continuous development of neural networks for computer vision tasks, more and more network architectures have achieved outstanding success. As one of the most advanced neural network architectures, DenseNet shortcuts all feature maps to solve the model depth problem. Although this network architecture has excellent accuracy with low parameters, it requires an excessive inference time. To solve this problem, HarDNet reduces the connections between the feature maps, making the remaining connections resemble harmonic waves. However, this compression method may result in a decrease in the model accuracy and an increase in the parameters and model size. This network architecture may reduce the memory access time, but its overall performance can still be improved. Therefore, we propose a new network architecture, ThreshNet, using a threshold mechanism to further optimize the connection method. Different numbers of connections for different convolution layers are discarded to accelerate the inference of the network. The proposed network has been evaluated with image classification using CIFAR 10 and SVHN datasets under platforms of NVIDIA RTX 3050 and Raspberry Pi 4. The experimental results show that, compared with HarDNet68, GhostNet, MobileNetV2, ShuffleNet, and EfficientNet, the inference time of the proposed ThreshNet79 is 5%, 9%, 10%, 18%, and 20% faster, respectively. The number of parameters of ThreshNet95 is 55% less than that of HarDNet85. The new model compression and model acceleration methods can speed up the inference time, enabling network models to operate on mobile devices.
Federated Learning (FL) has recently made significant progress as a new machine learning paradigm for privacy protection. Due to the high communication cost of traditional FL, one-shot federated learning is gaining popularity as a way to reduce communication cost between clients and the server. Most of the existing one-shot FL methods are based on Knowledge Distillation; however, {distillation based approach requires an extra training phase and depends on publicly available data sets or generated pseudo samples.} In this work, we consider a novel and challenging cross-silo setting: performing a single round of parameter aggregation on the local models without server-side training. In this setting, we propose an effective algorithm for Model Aggregation via Exploring Common Harmonized Optima (MA-Echo), which iteratively updates the parameters of all local models to bring them close to a common low-loss area on the loss surface, without harming performance on their own data sets at the same time. Compared to the existing methods, MA-Echo can work well even in extremely non-identical data distribution settings where the support categories of each local model have no overlapped labels with those of the others. We conduct extensive experiments on two popular image classification data sets to compare the proposed method with existing methods and demonstrate the effectiveness of MA-Echo, which clearly outperforms the state-of-the-arts. The source code can be accessed in \url{https://github.com/FudanVI/MAEcho}.
Feature representation learning is the key recipe for learning-based Multi-View Stereo (MVS). As the common feature extractor of learning-based MVS, vanilla Feature Pyramid Networks (FPNs) suffer from discouraged feature representations for reflection and texture-less areas, which limits the generalization of MVS. Even FPNs worked with pre-trained Convolutional Neural Networks (CNNs) fail to tackle these issues. On the other hand, Vision Transformers (ViTs) have achieved prominent success in many 2D vision tasks. Thus we ask whether ViTs can facilitate feature learning in MVS? In this paper, we propose a pre-trained ViT enhanced MVS network called MVSFormer, which can learn more reliable feature representations benefited by informative priors from ViT. The finetuned MVSFormer with hierarchical ViTs of efficient attention mechanisms can achieve prominent improvement based on FPNs. Besides, the alternative MVSFormer with frozen ViT weights is further proposed. This largely alleviates the training cost with competitive performance strengthened by the attention map from the self-distillation pre-training. MVSFormer can be generalized to various input resolutions with efficient multi-scale training strengthened by gradient accumulation. Moreover, we discuss the merits and drawbacks of classification and regression-based MVS methods, and further propose to unify them with a temperature-based strategy. MVSFormer achieves state-of-the-art performance on the DTU dataset. Particularly, MVSFormer ranks as Top-1 on both intermediate and advanced sets of the highly competitive Tanks-and-Temples leaderboard.
Colonoscopy is widely recognised as the gold standard procedure for the early detection of colorectal cancer (CRC). Segmentation is valuable for two significant clinical applications, namely lesion detection and classification, providing means to improve accuracy and robustness. The manual segmentation of polyps in colonoscopy images is time-consuming. As a result, the use of deep learning (DL) for automation of polyp segmentation has become important. However, DL-based solutions can be vulnerable to overfitting and the resulting inability to generalise to images captured by different colonoscopes. Recent transformer-based architectures for semantic segmentation both achieve higher performance and generalise better than alternatives, however typically predict a segmentation map of $\frac{h}{4}\times\frac{w}{4}$ spatial dimensions for a $h\times w$ input image. To this end, we propose a new architecture for full-size segmentation which leverages the strengths of a transformer in extracting the most important features for segmentation in a primary branch, while compensating for its limitations in full-size prediction with a secondary fully convolutional branch. The resulting features from both branches are then fused for final prediction of a $h\times w$ segmentation map. We demonstrate our method's state-of-the-art performance with respect to the mDice, mIoU, mPrecision, and mRecall metrics, on both the Kvasir-SEG and CVC-ClinicDB dataset benchmarks. Additionally, we train the model on each of these datasets and evaluate on the other to demonstrate its superior generalisation performance.