The conservation of stone-based cultural heritage sites is a critical concern for preserving cultural and historical landmarks. With the advent of Large Multimodal Models, as GPT-4omni (OpenAI), Claude 3 Opus (Anthropic) and Gemini 1.5 Pro (Google), it is becoming increasingly important to define the operational capabilities of these models. In this work, we systematically evaluate the abilities of the main foundational multimodal models to recognise and classify anomalies and deterioration patterns of the stone elements that are useful in the practice of conservation and restoration of world heritage. After defining a taxonomy of the main stone deterioration patterns and anomalies, we asked the foundational models to identify a curated selection of 354 highly representative images of stone-built heritage, offering them a careful selection of labels to choose from. The result, which varies depending on the type of pattern, allowed us to identify the strengths and weaknesses of these models in the field of heritage conservation and restoration.
Few-shot image classification involves classifying images using very few training examples. Recent vision foundation models show excellent few-shot transfer abilities, but are large and slow at inference. Using knowledge distillation, the capabilities of high-performing but slow models can be transferred to tiny, efficient models. However, common distillation methods require a large set of unlabeled data, which is not available in the few-shot setting. To overcome this lack of data, there has been a recent interest in using synthetic data. We expand on this work by presenting a novel diffusion model inversion technique (TINT) combining the diversity of textual inversion with the specificity of null-text inversion. Using this method in a few-shot distillation pipeline leads to state-of-the-art accuracy among small student models on popular benchmarks, while being significantly faster than prior work. This allows us to push even tiny models to high accuracy using only a tiny application-specific dataset, albeit relying on extra data for pre-training. Popular few-shot benchmarks involve evaluation over a large number of episodes, which is computationally cumbersome for methods involving synthetic data generation. Therefore, we also present a theoretical analysis on how the variance of the accuracy estimator depends on the number of episodes and query examples, and use these results to lower the computational effort required for method evaluation. In addition, to further motivate the use of generative models in few-shot distillation, we demonstrate that our method performs better compared to training on real data mined from the dataset used to train the diffusion model. Source code will be made available at https://github.com/pixwse/tiny2.
Tsetlin machines (TMs) have been successful in several application domains, operating with high efficiency on Boolean representations of the input data. However, Booleanizing complex data structures such as sequences, graphs, images, signal spectra, chemical compounds, and natural language is not trivial. In this paper, we propose a hypervector (HV) based method for expressing arbitrarily large sets of concepts associated with any input data. Using a hyperdimensional space to build vectors drastically expands the capacity and flexibility of the TM. We demonstrate how images, chemical compounds, and natural language text are encoded according to the proposed method, and how the resulting HV-powered TM can achieve significantly higher accuracy and faster learning on well-known benchmarks. Our results open up a new research direction for TMs, namely how to expand and exploit the benefits of operating in hyperspace, including new booleanization strategies, optimization of TM inference and learning, as well as new TM applications.
Deep model fusion is an emerging technique that unifies the predictions or parameters of several deep neural networks into a single model in a cost-effective and data-efficient manner. This enables the unified model to take advantage of the original models' strengths, potentially exceeding their performance. Although a variety of deep model fusion techniques have been introduced, their evaluations tend to be inconsistent and often inadequate to validate their effectiveness and robustness against distribution shifts. To address this issue, we introduce FusionBench, which is the first comprehensive benchmark dedicated to deep model fusion. FusionBench covers a wide range of tasks, including open-vocabulary image classification, text classification, and text-to-text generation. Each category includes up to eight tasks with corresponding task-specific models, featuring both full fine-tuning and LoRA fine-tuning, as well as models of different sizes, to ensure fair and balanced comparisons of various multi-task model fusion techniques across different tasks, model scales, and fine-tuning strategies. We implement and evaluate a broad spectrum of deep model fusion techniques. These techniques range from model ensemble methods, which combine the predictions to improve the overall performance, to model merging, which integrates different models into a single one, and model mixing methods, which upscale or recombine the components of the original models. FusionBench now contains 26 distinct tasks, 74 fine-tuned models, and 16 fusion techniques, and we are committed to consistently expanding the benchmark with more tasks, models, and fusion techniques. In addition, we offer a well-documented set of resources and guidelines to aid researchers in understanding and replicating the benchmark results. Homepage https://github.com/tanganke/fusion_bench
In this paper we address image classification tasks leveraging knowledge encoded in Large Multimodal Models (LMMs). More specifically, we use the MiniGPT-4 model to extract semantic descriptions for the images, in a multimodal prompting fashion. In the current literature, vision language models such as CLIP, among other approaches, are utilized as feature extractors, using only the image encoder, for solving image classification tasks. In this paper, we propose to additionally use the text encoder to obtain the text embeddings corresponding to the MiniGPT-4-generated semantic descriptions. Thus, we use both the image and text embeddings for solving the image classification task. The experimental evaluation on three datasets validates the improved classification performance achieved by exploiting LMM-based knowledge.
Deep neural networks exhibit remarkable performance, yet their black-box nature limits their utility in fields like healthcare where interpretability is crucial. Existing explainability approaches often sacrifice accuracy and lack quantifiable measures of prediction uncertainty. In this study, we introduce Conformal Prediction for Interpretable Neural Networks (CONFINE), a versatile framework that generates prediction sets with statistically robust uncertainty estimates instead of point predictions to enhance model transparency and reliability. CONFINE not only provides example-based explanations and confidence estimates for individual predictions but also boosts accuracy by up to 3.6%. We define a new metric, correct efficiency, to evaluate the fraction of prediction sets that contain precisely the correct label and show that CONFINE achieves correct efficiency of up to 3.3% higher than the original accuracy, matching or exceeding prior methods. CONFINE's marginal and class-conditional coverages attest to its validity across tasks spanning medical image classification to language understanding. Being adaptable to any pre-trained classifier, CONFINE marks a significant advance towards transparent and trustworthy deep learning applications in critical domains.
Federated learning (FL) systems enable multiple clients to train a machine learning model iteratively through synchronously exchanging the intermediate model weights with a single server. The scalability of such FL systems can be limited by two factors: server idle time due to synchronous communication and the risk of a single server becoming the bottleneck. In this paper, we propose a new FL architecture, to our knowledge, the first multi-server FL system that is entirely asynchronous, and therefore addresses these two limitations simultaneously. Our solution keeps both servers and clients continuously active. As in previous multi-server methods, clients interact solely with their nearest server, ensuring efficient update integration into the model. Differently, however, servers also periodically update each other asynchronously, and never postpone interactions with clients. We compare our solution to three representative baselines - FedAvg, FedAsync and HierFAVG - on the MNIST and CIFAR-10 image classification datasets and on the WikiText-2 language modeling dataset. Our solution converges to similar or higher accuracy levels than previous baselines and requires 61% less time to do so in geo-distributed settings.
In this research, we propose the first approach for integrating the Kolmogorov-Arnold Network (KAN) with various pre-trained Convolutional Neural Network (CNN) models for remote sensing (RS) scene classification tasks using the EuroSAT dataset. Our novel methodology, named KCN, aims to replace traditional Multi-Layer Perceptrons (MLPs) with KAN to enhance classification performance. We employed multiple CNN-based models, including VGG16, MobileNetV2, EfficientNet, ConvNeXt, ResNet101, and Vision Transformer (ViT), and evaluated their performance when paired with KAN. Our experiments demonstrated that KAN achieved high accuracy with fewer training epochs and parameters. Specifically, ConvNeXt paired with KAN showed the best performance, achieving 94% accuracy in the first epoch, which increased to 96% and remained consistent across subsequent epochs. The results indicated that KAN and MLP both achieved similar accuracy, with KAN performing slightly better in later epochs. By utilizing the EuroSAT dataset, we provided a robust testbed to investigate whether KAN is suitable for remote sensing classification tasks. Given that KAN is a novel algorithm, there is substantial capacity for further development and optimization, suggesting that KCN offers a promising alternative for efficient image analysis in the RS field.
Local learning offers an alternative to traditional end-to-end back-propagation in deep neural networks, significantly reducing GPU memory usage. While local learning has shown promise in image classification tasks, its application to other visual tasks remains limited. This limitation arises primarily from two factors: 1) architectures tailored for classification are often not transferable to other tasks, leading to a lack of reusability of task-specific knowledge; 2) the absence of cross-scale feature communication results in degraded performance in tasks such as object detection and super-resolution. To address these challenges, we propose the Memory-augmented Auxiliary Network (MAN), which introduces a simplified design principle and incorporates a feature bank to enhance cross-task adaptability and communication. This work represents the first successful application of local learning methods beyond classification, demonstrating that MAN not only conserves GPU memory but also achieves performance on par with end-to-end approaches across multiple datasets for various visual tasks.
The growing interest in satellite imagery has triggered the need for efficient mechanisms to extract valuable information from these vast data sources, providing deeper insights. Even though deep learning has shown significant progress in satellite image classification. Nevertheless, in the literature, only a few results can be found on weight initialization techniques. These techniques traditionally involve initializing the networks' weights before training on extensive datasets, distinct from fine-tuning the weights of pre-trained networks. In this study, a novel weight initialization method is proposed in the context of satellite image classification. The proposed weight initialization method is mathematically detailed during the forward and backward passes of the convolutional neural network (CNN) model. Extensive experiments are carried out using six real-world datasets. Comparative analyses with existing weight initialization techniques made on various well-known CNN models reveal that the proposed weight initialization technique outperforms the previous competitive techniques in classification accuracy. The complete code of the proposed technique, along with the obtained results, is available at https://github.com/WadiiBoulila/Weight-Initialization
Deep learning, particularly Convolutional Neural Networks (CNNs), has gained significant attention for its effectiveness in computer vision, especially in agricultural tasks. Recent advancements in instance segmentation have improved image classification accuracy. In this work, we introduce a comprehensive dataset for training neural networks to detect weeds and soy plants through instance segmentation. Our dataset covers various stages of soy growth, offering a chronological perspective on weed invasion's impact, with 1,000 meticulously annotated images. We also provide 6 state of the art models, trained in this dataset, that can understand and detect soy and weed in every stage of the plantation process. By using this dataset for weed and soy segmentation, we achieved a segmentation average precision of 79.1% and an average recall of 69.2% across all plant classes, with the YOLOv8X model. Moreover, the YOLOv8M model attained 78.7% mean average precision (mAp-50) in caruru weed segmentation, 69.7% in grassy weed segmentation, and 90.1% in soy plant segmentation.
Ultrasound is widely used in clinical practice due to its affordability, portability, and safety. However, current AI research often overlooks combined disease prediction and tissue segmentation. We propose UniUSNet, a universal framework for ultrasound image classification and segmentation. This model handles various ultrasound types, anatomical positions, and input formats, excelling in both segmentation and classification tasks. Trained on a comprehensive dataset with over 9.7K annotations from 7 distinct anatomical positions, our model matches state-of-the-art performance and surpasses single-dataset and ablated models. Zero-shot and fine-tuning experiments show strong generalization and adaptability with minimal fine-tuning. We plan to expand our dataset and refine the prompting mechanism, with model weights and code available at (https://github.com/Zehui-Lin/UniUSNet).
Few-shot action recognition (FSAR) aims to learn a model capable of identifying novel actions in videos using only a few examples. In assuming the base dataset seen during meta-training and novel dataset used for evaluation can come from different domains, cross-domain few-shot learning alleviates data collection and annotation costs required by methods with greater supervision and conventional (single-domain) few-shot methods. While this form of learning has been extensively studied for image classification, studies in cross-domain FSAR (CD-FSAR) are limited to proposing a model, rather than first understanding the cross-domain capabilities of existing models. To this end, we systematically evaluate existing state-of-the-art single-domain, transfer-based, and cross-domain FSAR methods on new cross-domain tasks with increasing difficulty, measured based on the domain shift between the base and novel set. Our empirical meta-analysis reveals a correlation between domain difference and downstream few-shot performance, and uncovers several important insights into which model aspects are effective for CD-FSAR and which need further development. Namely, we find that as the domain difference increases, the simple transfer-learning approach outperforms other methods by over 12 percentage points, and under these more challenging cross-domain settings, the specialised cross-domain model achieves the lowest performance. We also witness state-of-the-art single-domain FSAR models which use temporal alignment achieving similar or worse performance than earlier methods which do not, suggesting existing temporal alignment techniques fail to generalise on unseen domains. To the best of our knowledge, we are the first to systematically study the CD-FSAR problem in-depth. We hope the insights and challenges revealed in our study inspires and informs future work in these directions.
Recent advancements in machine learning, particularly in deep learning and object detection, have significantly improved performance in various tasks, including image classification and synthesis. However, challenges persist, particularly in acquiring labeled data that accurately represents specific use cases. In this work, we propose an automatic pipeline for generating synthetic image datasets using Stable Diffusion, an image synthesis model capable of producing highly realistic images. We leverage YOLOv8 for automatic bounding box detection and quality assessment of synthesized images. Our contributions include demonstrating the feasibility of training image classifiers solely on synthetic data, automating the image generation pipeline, and describing the computational requirements for our approach. We evaluate the usability of different modes of Stable Diffusion and achieve a classification accuracy of 75%.
Self-supervised learning (SSL) has potential for effective representation learning in medical imaging, but the choice of data augmentation is critical and domain-specific. It remains uncertain if general augmentation policies suit surgical applications. In this work, we automate the search for suitable augmentation policies through a new method called Dimensionality Driven Augmentation Search (DDA). DDA leverages the local dimensionality of deep representations as a proxy target, and differentiably searches for suitable data augmentation policies in contrastive learning. We demonstrate the effectiveness and efficiency of DDA in navigating a large search space and successfully identifying an appropriate data augmentation policy for laparoscopic surgery. We systematically evaluate DDA across three laparoscopic image classification and segmentation tasks, where it significantly improves over existing baselines. Furthermore, DDA's optimised set of augmentations provides insight into domain-specific dependencies when applying contrastive learning in medical applications. For example, while hue is an effective augmentation for natural images, it is not advantageous for laparoscopic images.
The Tsetlin Machine (TM) has achieved competitive results on several image classification benchmarks, including MNIST, K-MNIST, F-MNIST, and CIFAR-2. However, color image classification is arguably still in its infancy for TMs, with CIFAR-10 being a focal point for tracking progress. Over the past few years, TM's CIFAR-10 accuracy has increased from around 61% in 2020 to 75.1% in 2023 with the introduction of Drop Clause. In this paper, we leverage the recently proposed TM Composites architecture and introduce a range of TM Specialists that use various image processing techniques. These include Canny edge detection, Histogram of Oriented Gradients, adaptive mean thresholding, adaptive Gaussian thresholding, Otsu's thresholding, color thermometers, and adaptive color thermometers. In addition, we conduct a rigorous hyperparameter search, where we uncover optimal hyperparameters for several of the TM Specialists. The result is a toolbox that provides new state-of-the-art results on CIFAR-10 for TMs with an accuracy of 82.8%. In conclusion, our toolbox of TM Specialists forms a foundation for new TM applications and a landmark for further research on TM Composites in image analysis.
In the field of whole slide image (WSI) classification, multiple instance learning (MIL) serves as a promising approach, commonly decoupled into feature extraction and aggregation. In this paradigm, our observation reveals that discriminative embeddings are crucial for aggregation to the final prediction. Among all feature updating strategies, task-oriented ones can capture characteristics specifically for certain tasks. However, they can be prone to overfitting and contaminated by samples assigned with noisy labels. To address this issue, we propose a heuristic clustering-driven feature fine-tuning method (HC-FT) to enhance the performance of multiple instance learning by providing purified positive and hard negative samples. Our method first employs a well-trained MIL model to evaluate the confidence of patches. Then, patches with high confidence are marked as positive samples, while the remaining patches are used to identify crucial negative samples. After two rounds of heuristic clustering and selection, purified positive and hard negative samples are obtained to facilitate feature fine-tuning. The proposed method is evaluated on both CAMELYON16 and BRACS datasets, achieving an AUC of 97.13% and 85.85%, respectively, consistently outperforming all compared methods.
The state space models, employing recursively propagated features, demonstrate strong representation capabilities comparable to Transformer models and superior efficiency. However, constrained by the inherent geometric constraints of sequences, it still falls short in modeling long-range dependencies. To address this issue, we propose the GrootVL network, which first dynamically generates a tree topology based on spatial relationships and input features. Then, feature propagation is performed based on this graph, thereby breaking the original sequence constraints to achieve stronger representation capabilities. Additionally, we introduce a linear complexity dynamic programming algorithm to enhance long-range interactions without increasing computational cost. GrootVL is a versatile multimodal framework that can be applied to both visual and textual tasks. Extensive experiments demonstrate that our method significantly outperforms existing structured state space models on image classification, object detection and segmentation. Besides, by fine-tuning large language models, our approach achieves consistent improvements in multiple textual tasks at minor training cost.
Contrastive Analysis (CA) regards the problem of identifying patterns in images that allow distinguishing between a background (BG) dataset (i.e. healthy subjects) and a target (TG) dataset (i.e. unhealthy subjects). Recent works on this topic rely on variational autoencoders (VAE) or contrastive learning strategies to learn the patterns that separate TG samples from BG samples in a supervised manner. However, the dependency on target (unhealthy) samples can be challenging in medical scenarios due to their limited availability. Also, the blurred reconstructions of VAEs lack utility and interpretability. In this work, we redefine the CA task by employing a self-supervised contrastive encoder to learn a latent representation encoding only common patterns from input images, using samples exclusively from the BG dataset during training, and approximating the distribution of the target patterns by leveraging data augmentation techniques. Subsequently, we exploit state-of-the-art generative methods, i.e. diffusion models, conditioned on the learned latent representation to produce a realistic (healthy) version of the input image encoding solely the common patterns. Thorough validation on a facial image dataset and experiments across three brain MRI datasets demonstrate that conditioning the generative process of state-of-the-art generative methods with the latent representation from our self-supervised contrastive encoder yields improvements in the generated image quality and in the accuracy of image classification. The code is available at https://github.com/CristianoPatricio/unsupervised-contrastive-cond-diff.
Recent advancements in generative AI have introduced novel prospects and practical implementations. Especially diffusion models show their strength in generating diverse and, at the same time, realistic features, positioning them well for generating counterfactual explanations for computer vision models. Answering "what if" questions of what needs to change to make an image classifier change its prediction, counterfactual explanations align well with human understanding and consequently help in making model behavior more comprehensible. Current methods succeed in generating authentic counterfactuals, but lack transparency as feature changes are not directly perceivable. To address this limitation, we introduce Concept-guided Latent Diffusion Counterfactual Explanations (CoLa-DCE). CoLa-DCE generates concept-guided counterfactuals for any classifier with a high degree of control regarding concept selection and spatial conditioning. The counterfactuals comprise an increased granularity through minimal feature changes. The reference feature visualization ensures better comprehensibility, while the feature localization provides increased transparency of "where" changed "what". We demonstrate the advantages of our approach in minimality and comprehensibility across multiple image classification models and datasets and provide insights into how our CoLa-DCE explanations help comprehend model errors like misclassification cases.