Due to the depth degradation effect in residual connections, many efficient Vision Transformers models that rely on stacking layers for information exchange often fail to form sufficient information mixing, leading to unnatural visual perception. To address this issue, in this paper, we propose Aggregated Attention, a biomimetic design-based token mixer that simulates biological foveal vision and continuous eye movement while enabling each token on the feature map to have a global perception. Furthermore, we incorporate learnable tokens that interact with conventional queries and keys, which further diversifies the generation of affinity matrices beyond merely relying on the similarity between queries and keys. Our approach does not rely on stacking for information exchange, thus effectively avoiding depth degradation and achieving natural visual perception. Additionally, we propose Convolutional GLU, a channel mixer that bridges the gap between GLU and SE mechanism, which empowers each token to have channel attention based on its nearest neighbor image features, enhancing local modeling capability and model robustness. We combine aggregated attention and convolutional GLU to create a new visual backbone called TransNeXt. Extensive experiments demonstrate that our TransNeXt achieves state-of-the-art performance across multiple model sizes. At a resolution of $224^2$, TransNeXt-Tiny attains an ImageNet accuracy of 84.0%, surpassing ConvNeXt-B with 69% fewer parameters. Our TransNeXt-Base achieves an ImageNet accuracy of 86.2% and an ImageNet-A accuracy of 61.6% at a resolution of $384^2$, a COCO object detection mAP of 57.1, and an ADE20K semantic segmentation mIoU of 54.7.
With the availability of powerful text-to-image diffusion models, recent works have explored the use of synthetic data to improve image classification performances. These works show that it can effectively augment or even replace real data. In this work, we investigate how synthetic data can benefit few-shot class-agnostic counting. This requires to generate images that correspond to a given input number of objects. However, text-to-image models struggle to grasp the notion of count. We propose to rely on a double conditioning of Stable Diffusion with both a prompt and a density map in order to augment a training dataset for few-shot counting. Due to the small dataset size, the fine-tuned model tends to generate images close to the training images. We propose to enhance the diversity of synthesized images by exchanging captions between images thus creating unseen configurations of object types and spatial layout. Our experiments show that our diversified generation strategy significantly improves the counting accuracy of two recent and performing few-shot counting models on FSC147 and CARPK.
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).
Recently, Mix-style data augmentation methods (e.g., Mixup and CutMix) have shown promising performance in various visual tasks. However, these methods are primarily designed for single-label images, ignoring the considerable discrepancies between single- and multi-label images, i.e., a multi-label image involves multiple co-occurred categories and fickle object scales. On the other hand, previous multi-label image classification (MLIC) methods tend to design elaborate models, bringing expensive computation. In this paper, we introduce a simple but effective augmentation strategy for multi-label image classification, namely SpliceMix. The "splice" in our method is two-fold: 1) Each mixed image is a splice of several downsampled images in the form of a grid, where the semantics of images attending to mixing are blended without object deficiencies for alleviating co-occurred bias; 2) We splice mixed images and the original mini-batch to form a new SpliceMixed mini-batch, which allows an image with different scales to contribute to training together. Furthermore, such splice in our SpliceMixed mini-batch enables interactions between mixed images and original regular images. We also offer a simple and non-parametric extension based on consistency learning (SpliceMix-CL) to show the flexible extensibility of our SpliceMix. Extensive experiments on various tasks demonstrate that only using SpliceMix with a baseline model (e.g., ResNet) achieves better performance than state-of-the-art methods. Moreover, the generalizability of our SpliceMix is further validated by the improvements in current MLIC methods when married with our SpliceMix. The code is available at https://github.com/zuiran/SpliceMix.
Vision Transformers (ViTs) have been shown to enhance visual recognition through modeling long-range dependencies with multi-head self-attention (MHSA), which is typically formulated as Query-Key-Value computation. However, the attention map generated from the Query and Key captures only token-to-token correlations at one single granularity. In this paper, we argue that self-attention should have a more comprehensive mechanism to capture correlations among tokens and groups (i.e., multiple adjacent tokens) for higher representational capacity. Thereby, we propose Group-Mix Attention (GMA) as an advanced replacement for traditional self-attention, which can simultaneously capture token-to-token, token-to-group, and group-to-group correlations with various group sizes. To this end, GMA splits the Query, Key, and Value into segments uniformly and performs different group aggregations to generate group proxies. The attention map is computed based on the mixtures of tokens and group proxies and used to re-combine the tokens and groups in Value. Based on GMA, we introduce a powerful backbone, namely GroupMixFormer, which achieves state-of-the-art performance in image classification, object detection, and semantic segmentation with fewer parameters than existing models. For instance, GroupMixFormer-L (with 70.3M parameters and 384^2 input) attains 86.2% Top-1 accuracy on ImageNet-1K without external data, while GroupMixFormer-B (with 45.8M parameters) attains 51.2% mIoU on ADE20K.
Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent delta-tuning methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art fails to exceed the upper limit of full fine-tuning on challenging tasks like instance segmentation and semantic segmentation. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona) tuning, a novel adapter-based tuning method. First, we introduce multiple vision-friendly filters into the adapter to enhance its ability to process visual signals, while previous methods mainly rely on language-friendly linear filters. Second, we add the scaled normalization layer in the adapter to regulate the distribution of input features for visual filters. To fully demonstrate the practicality and generality of Mona, we conduct experiments on multiple representative visual tasks, including instance segmentation on COCO, semantic segmentation on ADE20K, object detection on Pascal VOC, and image classification on several common datasets. Exciting results illustrate that Mona surpasses full fine-tuning on all these tasks and is the only delta-tuning method outperforming full fine-tuning on instance segmentation and semantic segmentation tasks. For example, Mona achieves a 1% performance gain on the COCO dataset compared to full fine-tuning. Comprehensive results suggest that Mona-tuning is more suitable for retaining and utilizing the capabilities of pre-trained models than full fine-tuning. The code will be released at https://github.com/Leiyi-Hu/mona.
The dissection of hyperspectral images into intrinsic components through hyperspectral intrinsic image decomposition (HIID) enhances the interpretability of hyperspectral data, providing a foundation for more accurate classification outcomes. However, the classification performance of HIID is constrained by the model's representational ability. To address this limitation, this study rethinks hyperspectral intrinsic image decomposition for classification tasks by introducing deep feature embedding. The proposed framework, HyperDID, incorporates the Environmental Feature Module (EFM) and Categorical Feature Module (CFM) to extract intrinsic features. Additionally, a Feature Discrimination Module (FDM) is introduced to separate environment-related and category-related features. Experimental results across three commonly used datasets validate the effectiveness of HyperDID in improving hyperspectral image classification performance. This novel approach holds promise for advancing the capabilities of hyperspectral image analysis by leveraging deep feature embedding principles. The implementation of the proposed method could be accessed soon at https://github.com/shendu-sw/HyperDID for the sake of reproducibility.
Large-kernel convolutional neural networks (ConvNets) have recently received extensive research attention, but two unresolved and critical issues demand further investigation. 1) The architectures of existing large-kernel ConvNets largely follow the design principles of conventional ConvNets or transformers, while the architectural design for large-kernel ConvNets remains under-addressed. 2) As transformers have dominated multiple modalities, it remains to be investigated whether ConvNets also have a strong universal perception ability in domains beyond vision. In this paper, we contribute from two aspects. 1) We propose four architectural guidelines for designing large-kernel ConvNets, the core of which is to exploit the essential characteristics of large kernels that distinguish them from small kernels - they can see wide without going deep. Following such guidelines, our proposed large-kernel ConvNet shows leading performance in image recognition (ImageNet accuracy of 88.0%, ADE20K mIoU of 55.6%, and COCO box AP of 56.4%), demonstrating better performance and higher speed than the recent powerful competitors. 2) We discover large kernels are the key to unlocking the exceptional performance of ConvNets in domains where they were originally not proficient. With certain modality-related preprocessing approaches, the proposed model achieves state-of-the-art performance on time-series forecasting and audio recognition tasks even without modality-specific customization to the architecture. All the code and models are publicly available on GitHub and Huggingface.
Vision-language models (VLMs) like CLIP have demonstrated remarkable applicability across a variety of downstream tasks, including zero-shot image classification. Recently, the use of prompts or adapters for efficient transfer learning (ETL) has gained significant attention for effectively adapting to downstream tasks. However, previous studies have overlooked the challenge of varying transfer difficulty of downstream tasks. In this paper, we empirically analyze how each ETL method behaves with respect to transfer difficulty. Our observations indicate that utilizing vision prompts and text adapters is crucial for adaptability and generalizability in domains with high difficulty. Also, by applying an adaptive ensemble approach that integrates task-adapted VLMs with pre-trained VLMs and strategically leverages more general knowledge in low-difficulty and less in high-difficulty domains, we consistently enhance performance across both types of domains. Based on these observations, we propose an adaptive ensemble method that combines visual prompts and text adapters with pre-trained VLMs, tailored by transfer difficulty, to achieve optimal performance for any target domain. Upon experimenting with extensive benchmarks, our method consistently outperforms all baselines, particularly on unseen tasks, demonstrating its effectiveness.
We propose an approach to do learning in Gaussian factor graphs. We treat all relevant quantities (inputs, outputs, parameters, latents) as random variables in a graphical model, and view both training and prediction as inference problems with different observed nodes. Our experiments show that these problems can be efficiently solved with belief propagation (BP), whose updates are inherently local, presenting exciting opportunities for distributed and asynchronous training. Our approach can be scaled to deep networks and provides a natural means to do continual learning: use the BP-estimated parameter marginals of the current task as parameter priors for the next. On a video denoising task we demonstrate the benefit of learnable parameters over a classical factor graph approach and we show encouraging performance of deep factor graphs for continual image classification.
Although current data augmentation methods are successful to alleviate the data insufficiency, conventional augmentation are primarily intra-domain while advanced generative adversarial networks (GANs) generate images remaining uncertain, particularly in small-scale datasets. In this paper, we propose a parameterized GAN (ParaGAN) that effectively controls the changes of synthetic samples among domains and highlights the attention regions for downstream classification. Specifically, ParaGAN incorporates projection distance parameters in cyclic projection and projects the source images to the decision boundary to obtain the class-difference maps. Our experiments show that ParaGAN can consistently outperform the existing augmentation methods with explainable classification on two small-scale medical datasets.
Federated Learning (FL) is a collaborative training paradigm that allows for privacy-preserving learning of cross-institutional models by eliminating the exchange of sensitive data and instead relying on the exchange of model parameters between the clients and a server. Despite individual studies on how client models are aggregated, and, more recently, on the benefits of ImageNet pre-training, there is a lack of understanding of the effect the architecture chosen for the federation has, and of how the aforementioned elements interconnect. To this end, we conduct the first joint ARchitecture-Initialization-Aggregation study and benchmark ARIAs across a range of medical image classification tasks. We find that, contrary to current practices, ARIA elements have to be chosen together to achieve the best possible performance. Our results also shed light on good choices for each element depending on the task, the effect of normalisation layers, and the utility of SSL pre-training, pointing to potential directions for designing FL-specific architectures and training pipelines.
Deep neural networks (DNNs) exhibit superior performance in various machine learning tasks, e.g., image classification, speech recognition, biometric recognition, object detection, etc. However, it is essential to analyze their sensitivity to parameter perturbations before deploying them in real-world applications. In this work, we assess the sensitivity of DNNs against perturbations to their weight and bias parameters. The sensitivity analysis involves three DNN architectures (VGG, ResNet, and DenseNet), three types of parameter perturbations (Gaussian noise, weight zeroing, and weight scaling), and two settings (entire network and layer-wise). We perform experiments in the context of iris presentation attack detection and evaluate on two publicly available datasets: LivDet-Iris-2017 and LivDet-Iris-2020. Based on the sensitivity analysis, we propose improved models simply by perturbing parameters of the network without undergoing training. We further combine these perturbed models at the score-level and at the parameter-level to improve the performance over the original model. The ensemble at the parameter-level shows an average improvement of 43.58% on the LivDet-Iris-2017 dataset and 9.25% on the LivDet-Iris-2020 dataset. The source code is available at https://github.com/redwankarimsony/WeightPerturbation-MSU.
Vision-language models (VLMs) offer a promising paradigm for image classification by comparing the similarity between images and class embeddings. A critical challenge lies in crafting precise textual representations for class names. While previous studies have leveraged recent advancements in large language models (LLMs) to enhance these descriptors, their outputs often suffer from ambiguity and inaccuracy. We attribute this to two primary factors: 1) the reliance on single-turn textual interactions with LLMs, leading to a mismatch between generated text and visual concepts for VLMs; 2) the oversight of the inter-class relationships, resulting in descriptors that fail to differentiate similar classes effectively. In this paper, we propose a novel framework that integrates LLMs and VLMs to find the optimal class descriptors. Our training-free approach develops an LLM-based agent with an evolutionary optimization strategy to iteratively refine class descriptors. We demonstrate our optimized descriptors are of high quality which effectively improves classification accuracy on a wide range of benchmarks. Additionally, these descriptors offer explainable and robust features, boosting performance across various backbone models and complementing fine-tuning-based methods.
Weakly supervised whole slide image classification is a key task in computational pathology, which involves predicting a slide-level label from a set of image patches constituting the slide. Constructing models to solve this task involves multiple design choices, often made without robust empirical or conclusive theoretical justification. To address this, we conduct a comprehensive benchmarking of feature extractors to answer three critical questions: 1) Is stain normalisation still a necessary preprocessing step? 2) Which feature extractors are best for downstream slide-level classification? 3) How does magnification affect downstream performance? Our study constitutes the most comprehensive evaluation of publicly available pathology feature extractors to date, involving more than 10,000 training runs across 14 feature extractors, 9 tasks, 5 datasets, 3 downstream architectures, 2 levels of magnification, and various preprocessing setups. Our findings challenge existing assumptions: 1) We observe empirically, and by analysing the latent space, that skipping stain normalisation and image augmentations does not degrade performance, while significantly reducing memory and computational demands. 2) We develop a novel evaluation metric to compare relative downstream performance, and show that the choice of feature extractor is the most consequential factor for downstream performance. 3) We find that lower-magnification slides are sufficient for accurate slide-level classification. Contrary to previous patch-level benchmarking studies, our approach emphasises clinical relevance by focusing on slide-level biomarker prediction tasks in a weakly supervised setting with external validation cohorts. Our findings stand to streamline digital pathology workflows by minimising preprocessing needs and informing the selection of feature extractors.
Convolutional Neural Networks (CNNs) have revolutionized image classification by extracting spatial features and enabling state-of-the-art accuracy in vision-based tasks. The squeeze and excitation network proposed module gathers channelwise representations of the input. Multilayer perceptrons (MLP) learn global representation from the data and in most image classification models used to learn extracted features of the image. In this paper, we introduce a novel aggregated multilayer perceptron, a multi-branch dense layer, within the Squeeze excitation residual module designed to surpass the performance of existing architectures. Our approach leverages a combination of squeeze excitation network module with dense layers. This fusion enhances the network's ability to capture channel-wise patterns and have global knowledge, leading to a better feature representation. This proposed model has a negligible increase in parameters when compared to SENet. We conduct extensive experiments on benchmark datasets to validate the model and compare them with established architectures. Experimental results demonstrate a remarkable increase in the classification accuracy of the proposed model.
Object detection models are commonly used for people counting (and localization) in many applications but require a dataset with costly bounding box annotations for training. Given the importance of privacy in people counting, these models rely more and more on infrared images, making the task even harder. In this paper, we explore how weaker levels of supervision can affect the performance of deep person counting architectures for image classification and point-level localization. Our experiments indicate that counting people using a CNN Image-Level model achieves competitive results with YOLO detectors and point-level models, yet provides a higher frame rate and a similar amount of model parameters.
Regularization is a critical component in deep learning. The most commonly used approach, weight decay, applies a constant penalty coefficient uniformly across all parameters. This may be overly restrictive for some parameters, while insufficient for others. To address this, we present Constrained Parameter Regularization (CPR) as an alternative to traditional weight decay. Unlike the uniform application of a single penalty, CPR enforces an upper bound on a statistical measure, such as the L2-norm, of individual parameter matrices. Consequently, learning becomes a constraint optimization problem, which we tackle using an adaptation of the augmented Lagrangian method. CPR introduces only a minor runtime overhead and only requires setting an upper bound. We propose simple yet efficient mechanisms for initializing this bound, making CPR rely on no hyperparameter or one, akin to weight decay. Our empirical studies on computer vision and language modeling tasks demonstrate CPR's effectiveness. The results show that CPR can outperform traditional weight decay and increase performance in pre-training and fine-tuning.
Uncertainty in machine learning models is a timely and vast field of research. In supervised learning, uncertainty can already occur in the first stage of the training process, the annotation phase. This scenario is particularly evident when some instances cannot be definitively classified. In other words, there is inevitable ambiguity in the annotation step and hence, not necessarily a "ground truth" associated with each instance. The main idea of this work is to drop the assumption of a ground truth label and instead embed the annotations into a multidimensional space. This embedding is derived from the empirical distribution of annotations in a Bayesian setup, modeled via a Dirichlet-Multinomial framework. We estimate the model parameters and posteriors using a stochastic Expectation Maximization algorithm with Markov Chain Monte Carlo steps. The methods developed in this paper readily extend to various situations where multiple annotators independently label instances. To showcase the generality of the proposed approach, we apply our approach to three benchmark datasets for image classification and Natural Language Inference. Besides the embeddings, we can investigate the resulting correlation matrices, which reflect the semantic similarities of the original classes very well for all three exemplary datasets.
Although deep learning-based segmentation models have achieved impressive performance on public benchmarks, generalizing well to unseen environments remains a major challenge. To improve the model's generalization ability to the new domain during evaluation, the test-time training (TTT) is a challenging paradigm that adapts the source-pretrained model in an online fashion. Early efforts on TTT mainly focus on the image classification task. Directly extending these methods to semantic segmentation easily experiences unstable adaption due to segmentation's inherent characteristics, such as extreme class imbalance and complex decision spaces. To stabilize the adaptation process, we introduce contrastive loss (CL), known for its capability to learn robust and generalized representations. Nevertheless, the traditional CL operates in the representation space and cannot directly enhance predictions. In this paper, we resolve this limitation by adapting the CL to the output space, employing a high temperature, and simplifying the formulation, resulting in a straightforward yet effective loss function called Output Contrastive Loss (OCL). Our comprehensive experiments validate the efficacy of our approach across diverse evaluation scenarios. Notably, our method excels even when applied to models initially pre-trained using domain adaptation methods on test domain data, showcasing its resilience and adaptability.\footnote{Code and more information could be found at~ \url{https://github.com/dazhangyu123/OCL}}