Large language models are built on top of a transformer-based architecture to process textual inputs. For example, the LLaMA stands out among many open-source implementations. Can the same transformer be used to process 2D images? In this paper, we answer this question by unveiling a LLaMA-like vision transformer in plain and pyramid forms, termed VisionLLaMA, which is tailored for this purpose. VisionLLaMA is a unified and generic modelling framework for solving most vision tasks. We extensively evaluate its effectiveness using typical pre-training paradigms in a good portion of downstream tasks of image perception and especially image generation. In many cases, VisionLLaMA have exhibited substantial gains over the previous state-of-the-art vision transformers. We believe that VisionLLaMA can serve as a strong new baseline model for vision generation and understanding. Our code is released at https://github.com/Meituan-AutoML/VisionLLaMA.
Most existing weakly supervised semantic segmentation (WSSS) methods rely on Class Activation Mapping (CAM) to extract coarse class-specific localization maps using image-level labels. Prior works have commonly used an off-line heuristic thresholding process that combines the CAM maps with off-the-shelf saliency maps produced by a general pre-trained saliency model to produce more accurate pseudo-segmentation labels. We propose AuxSegNet+, a weakly supervised auxiliary learning framework to explore the rich information from these saliency maps and the significant inter-task correlation between saliency detection and semantic segmentation. In the proposed AuxSegNet+, saliency detection and multi-label image classification are used as auxiliary tasks to improve the primary task of semantic segmentation with only image-level ground-truth labels. We also propose a cross-task affinity learning mechanism to learn pixel-level affinities from the saliency and segmentation feature maps. In particular, we propose a cross-task dual-affinity learning module to learn both pairwise and unary affinities, which are used to enhance the task-specific features and predictions by aggregating both query-dependent and query-independent global context for both saliency detection and semantic segmentation. The learned cross-task pairwise affinity can also be used to refine and propagate CAM maps to provide better pseudo labels for both tasks. Iterative improvement of segmentation performance is enabled by cross-task affinity learning and pseudo-label updating. Extensive experiments demonstrate the effectiveness of the proposed approach with new state-of-the-art WSSS results on the challenging PASCAL VOC and MS COCO benchmarks.
The attention mechanism has gained significant recognition in the field of computer vision due to its ability to effectively enhance the performance of deep neural networks. However, existing methods often struggle to effectively utilize spatial information or, if they do, they come at the cost of reducing channel dimensions or increasing the complexity of neural networks. In order to address these limitations, this paper introduces an Efficient Local Attention (ELA) method that achieves substantial performance improvements with a simple structure. By analyzing the limitations of the Coordinate Attention method, we identify the lack of generalization ability in Batch Normalization, the adverse effects of dimension reduction on channel attention, and the complexity of attention generation process. To overcome these challenges, we propose the incorporation of 1D convolution and Group Normalization feature enhancement techniques. This approach enables accurate localization of regions of interest by efficiently encoding two 1D positional feature maps without the need for dimension reduction, while allowing for a lightweight implementation. We carefully design three hyperparameters in ELA, resulting in four different versions: ELA-T, ELA-B, ELA-S, and ELA-L, to cater to the specific requirements of different visual tasks such as image classification, object detection and sementic segmentation. ELA can be seamlessly integrated into deep CNN networks such as ResNet, MobileNet, and DeepLab. Extensive evaluations on the ImageNet, MSCOCO, and Pascal VOC datasets demonstrate the superiority of the proposed ELA module over current state-of-the-art methods in all three aforementioned visual tasks.
Vision-language foundation models have exhibited remarkable success across a multitude of downstream tasks due to their scalability on extensive image-text paired data. However, these models also display significant limitations when applied to downstream tasks, such as fine-grained image classification, as a result of ``decision shortcuts'' that hinder their generalization capabilities. In this work, we find that the CLIP model possesses a rich set of features, encompassing both \textit{desired invariant causal features} and \textit{undesired decision shortcuts}. Moreover, the underperformance of CLIP on downstream tasks originates from its inability to effectively utilize pre-trained features in accordance with specific task requirements. To address this challenge, we propose a simple yet effective method, Spurious Feature Eraser (SEraser), to alleviate the decision shortcuts by erasing the spurious features. Specifically, we introduce a test-time prompt tuning paradigm that optimizes a learnable prompt, thereby compelling the model to exploit invariant features while disregarding decision shortcuts during the inference phase. The proposed method effectively alleviates excessive dependence on potentially misleading spurious information. We conduct comparative analysis of the proposed method against various approaches which validates the significant superiority.
In this paper, we revisit techniques for uncertainty estimation within deep neural networks and consolidate a suite of techniques to enhance their reliability. Our investigation reveals that an integrated application of diverse techniques--spanning model regularization, classifier and optimization--substantially improves the accuracy of uncertainty predictions in image classification tasks. The synergistic effect of these techniques culminates in our novel SURE approach. We rigorously evaluate SURE against the benchmark of failure prediction, a critical testbed for uncertainty estimation efficacy. Our results showcase a consistently better performance than models that individually deploy each technique, across various datasets and model architectures. When applied to real-world challenges, such as data corruption, label noise, and long-tailed class distribution, SURE exhibits remarkable robustness, delivering results that are superior or on par with current state-of-the-art specialized methods. Particularly on Animal-10N and Food-101N for learning with noisy labels, SURE achieves state-of-the-art performance without any task-specific adjustments. This work not only sets a new benchmark for robust uncertainty estimation but also paves the way for its application in diverse, real-world scenarios where reliability is paramount. Our code is available at \url{https://yutingli0606.github.io/SURE/}.
Deep neural networks (DNNs) have achieved great breakthroughs in many fields such as image classification and natural language processing. However, the execution of DNNs needs to conduct massive numbers of multiply-accumulate (MAC) operations on hardware and thus incurs a large power consumption. To address this challenge, we propose a novel digital MAC design based on encoding. In this new design, the multipliers are replaced by simple logic gates to represent the results with a wide bit representation. The outputs of the new multipliers are added by bit-wise weighted accumulation and the accumulation results are compatible with existing computing platforms accelerating neural networks. Since the multiplication function is replaced by a simple logic representation, the critical paths in the resulting circuits become much shorter. Correspondingly, pipelining stages and intermediate registers used to store partial sums in the MAC array can be reduced, leading to a significantly smaller area as well as better power efficiency. The proposed design has been synthesized and verified by ResNet18- Cifar10, ResNet20-Cifar100, ResNet50-ImageNet, MobileNetV2-Cifar10, MobileNetV2-Cifar100, and EfficientNetB0-ImageNet. The experimental results confirmed the reduction of circuit area by up to 48.79% and the reduction of power consumption of executing DNNs by up to 64.41%, while the accuracy of the neural networks can still be well maintained. The open source code of this work can be found on GitHub with link https://github.com/Bo-Liu-TUM/EncodingNet/.
While standard Empirical Risk Minimization (ERM) training is proven effective for image classification on in-distribution data, it fails to perform well on out-of-distribution samples. One of the main sources of distribution shift for image classification is the compositional nature of images. Specifically, in addition to the main object or component(s) determining the label, some other image components usually exist, which may lead to the shift of input distribution between train and test environments. More importantly, these components may have spurious correlations with the label. To address this issue, we propose Decompose-and-Compose (DaC), which improves robustness to correlation shift by a compositional approach based on combining elements of images. Based on our observations, models trained with ERM usually highly attend to either the causal components or the components having a high spurious correlation with the label (especially in datapoints on which models have a high confidence). In fact, according to the amount of spurious correlation and the easiness of classification based on the causal or non-causal components, the model usually attends to one of these more (on samples with high confidence). Following this, we first try to identify the causal components of images using class activation maps of models trained with ERM. Afterward, we intervene on images by combining them and retraining the model on the augmented data, including the counterfactual ones. Along with its high interpretability, this work proposes a group-balancing method by intervening on images without requiring group labels or information regarding the spurious features during training. The method has an overall better worst group accuracy compared to previous methods with the same amount of supervision on the group labels in correlation shift.
The recently discovered Neural collapse (NC) phenomenon states that the last-layer weights of Deep Neural Networks (DNN), converge to the so-called Equiangular Tight Frame (ETF) simplex, at the terminal phase of their training. This ETF geometry is equivalent to vanishing within-class variability of the last layer activations. Inspired by NC properties, we explore in this paper the transferability of DNN models trained with their last layer weight fixed according to ETF. This enforces class separation by eliminating class covariance information, effectively providing implicit regularization. We show that DNN models trained with such a fixed classifier significantly improve transfer performance, particularly on out-of-domain datasets. On a broad range of fine-grained image classification datasets, our approach outperforms i) baseline methods that do not perform any covariance regularization (up to 22%), as well as ii) methods that explicitly whiten covariance of activations throughout training (up to 19%). Our findings suggest that DNNs trained with fixed ETF classifiers offer a powerful mechanism for improving transfer learning across domains.
While Neural Networks (NNs) have surpassed human accuracy in image classification on ImageNet, they often lack robustness against image corruption, i.e., corruption robustness. Yet such robustness is seemingly effortless for human perception. In this paper, we propose visually-continuous corruption robustness (VCR) -- an extension of corruption robustness to allow assessing it over the wide and continuous range of changes that correspond to the human perceptive quality (i.e., from the original image to the full distortion of all perceived visual information), along with two novel human-aware metrics for NN evaluation. To compare VCR of NNs with human perception, we conducted extensive experiments on 14 commonly used image corruptions with 7,718 human participants and state-of-the-art robust NN models with different training objectives (e.g., standard, adversarial, corruption robustness), different architectures (e.g., convolution NNs, vision transformers), and different amounts of training data augmentation. Our study showed that: 1) assessing robustness against continuous corruption can reveal insufficient robustness undetected by existing benchmarks; as a result, 2) the gap between NN and human robustness is larger than previously known; and finally, 3) some image corruptions have a similar impact on human perception, offering opportunities for more cost-effective robustness assessments. Our validation set with 14 image corruptions, human robustness data, and the evaluation code is provided as a toolbox and a benchmark.
Whole Slide Image (WSI) classification is often formulated as a Multiple Instance Learning (MIL) problem. Recently, Vision-Language Models (VLMs) have demonstrated remarkable performance in WSI classification. However, existing methods leverage coarse-grained pathogenetic descriptions for visual representation supervision, which are insufficient to capture the complex visual appearance of pathogenetic images, hindering the generalizability of models on diverse downstream tasks. Additionally, processing high-resolution WSIs can be computationally expensive. In this paper, we propose a novel "Fine-grained Visual-Semantic Interaction" (FiVE) framework for WSI classification. It is designed to enhance the model's generalizability by leveraging the interaction between localized visual patterns and fine-grained pathological semantics. Specifically, with meticulously designed queries, we start by utilizing a large language model to extract fine-grained pathological descriptions from various non-standardized raw reports. The output descriptions are then reconstructed into fine-grained labels used for training. By introducing a Task-specific Fine-grained Semantics (TFS) module, we enable prompts to capture crucial visual information in WSIs, which enhances representation learning and augments generalization capabilities significantly. Furthermore, given that pathological visual patterns are redundantly distributed across tissue slices, we sample a subset of visual instances during training. Our method demonstrates robust generalizability and strong transferability, dominantly outperforming the counterparts on the TCGA Lung Cancer dataset with at least 9.19% higher accuracy in few-shot experiments. The code is available at: https://github.com/ls1rius/WSI_FiVE.
In computer vision, a larger effective receptive field (ERF) is associated with better performance. While attention natively supports global context, its quadratic complexity limits its applicability to tasks that benefit from high-resolution input. In this work, we extend Hyena, a convolution-based attention replacement, from causal sequences to bidirectional data and two-dimensional image space. We scale Hyena's convolution kernels beyond the feature map size, up to 191$\times$191, to maximize ERF while maintaining sub-quadratic complexity in the number of pixels. We integrate our two-dimensional Hyena, HyenaPixel, and bidirectional Hyena into the MetaFormer framework. For image categorization, HyenaPixel and bidirectional Hyena achieve a competitive ImageNet-1k top-1 accuracy of 84.9% and 85.2%, respectively, with no additional training data, while outperforming other convolutional and large-kernel networks. Combining HyenaPixel with attention further improves accuracy. We attribute the success of bidirectional Hyena to learning the data-dependent geometric arrangement of pixels without a fixed neighborhood definition. Experimental results on downstream tasks suggest that HyenaPixel with large filters and a fixed neighborhood leads to better localization performance.
The increasing demand for automatic high-level image understanding, particularly in detecting abstract concepts (AC) within images, underscores the necessity for innovative and more interpretable approaches. These approaches need to harmonize traditional deep vision methods with the nuanced, context-dependent knowledge humans employ to interpret images at intricate semantic levels. In this work, we leverage situated perceptual knowledge of cultural images to enhance performance and interpretability in AC image classification. We automatically extract perceptual semantic units from images, which we then model and integrate into the ARTstract Knowledge Graph (AKG). This resource captures situated perceptual semantics gleaned from over 14,000 cultural images labeled with ACs. Additionally, we enhance the AKG with high-level linguistic frames. We compute KG embeddings and experiment with relative representations and hybrid approaches that fuse these embeddings with visual transformer embeddings. Finally, for interpretability, we conduct posthoc qualitative analyses by examining model similarities with training instances. Our results show that our hybrid KGE-ViT methods outperform existing techniques in AC image classification. The posthoc interpretability analyses reveal the visual transformer's proficiency in capturing pixel-level visual attributes, contrasting with our method's efficacy in representing more abstract and semantic scene elements. We demonstrate the synergy and complementarity between KGE embeddings' situated perceptual knowledge and deep visual model's sensory-perceptual understanding for AC image classification. This work suggests a strong potential of neuro-symbolic methods for knowledge integration and robust image representation for use in downstream intricate visual comprehension tasks. All the materials and code are available online.
In the rapidly evolving field of deep learning, the demand for models that are both expressive and computationally efficient has never been more critical. This paper introduces Orchid, a novel architecture designed to address the quadratic complexity of traditional attention mechanisms without compromising the ability to capture long-range dependencies and in-context learning. At the core of this architecture lies a new data-dependent global convolution layer, which contextually adapts its kernel conditioned on input sequence using a dedicated conditioning neural network. We design two simple conditioning networks that maintain shift equivariance in our data-dependent convolution operation. The dynamic nature of the proposed convolution kernel grants Orchid high expressivity while maintaining quasilinear scalability for long sequences. We evaluate the proposed model across multiple domains, including language modeling and image classification, to highlight its performance and generality. Our experiments demonstrate that this architecture not only outperforms traditional attention-based architectures such as BERT and Vision Transformers with smaller model sizes, but also extends the feasible sequence length beyond the limitations of the dense attention layers. This achievement represents a significant step towards more efficient and scalable deep learning models for sequence modeling.
Single-domain generalization aims to learn a model from single source domain data to achieve generalized performance on other unseen target domains. Existing works primarily focus on improving the generalization ability of static networks. However, static networks are unable to dynamically adapt to the diverse variations in different image scenes, leading to limited generalization capability. Different scenes exhibit varying levels of complexity, and the complexity of images further varies significantly in cross-domain scenarios. In this paper, we propose a dynamic object-centric perception network based on prompt learning, aiming to adapt to the variations in image complexity. Specifically, we propose an object-centric gating module based on prompt learning to focus attention on the object-centric features guided by the various scene prompts. Then, with the object-centric gating masks, the dynamic selective module dynamically selects highly correlated feature regions in both spatial and channel dimensions enabling the model to adaptively perceive object-centric relevant features, thereby enhancing the generalization capability. Extensive experiments were conducted on single-domain generalization tasks in image classification and object detection. The experimental results demonstrate that our approach outperforms state-of-the-art methods, which validates the effectiveness and generally of our proposed method.
In multimedia understanding tasks, corrupted samples pose a critical challenge, because when fed to machine learning models they lead to performance degradation. In the past, three groups of approaches have been proposed to handle noisy data: i) enhancer and denoiser modules to improve the quality of the noisy data, ii) data augmentation approaches, and iii) domain adaptation strategies. All the aforementioned approaches come with drawbacks that limit their applicability; the first has high computational costs and requires pairs of clean-corrupted data for training, while the others only allow deployment of the same task/network they were trained on (\ie, when upstream and downstream task/network are the same). In this paper, we propose SyMPIE to solve these shortcomings. To this end, we design a small, modular, and efficient (just 2GFLOPs to process a Full HD image) system to enhance input data for robust downstream multimedia understanding with minimal computational cost. Our SyMPIE is pre-trained on an upstream task/network that should not match the downstream ones and does not need paired clean-corrupted samples. Our key insight is that most input corruptions found in real-world tasks can be modeled through global operations on color channels of images or spatial filters with small kernels. We validate our approach on multiple datasets and tasks, such as image classification (on ImageNetC, ImageNetC-Bar, VizWiz, and a newly proposed mixed corruption benchmark named ImageNetC-mixed) and semantic segmentation (on Cityscapes, ACDC, and DarkZurich) with consistent improvements of about 5\% relative accuracy gain across the board. The code of our approach and the new ImageNetC-mixed benchmark will be made available upon publication.
In this paper, we present an empirical study on image recognition fairness, i.e., extreme class accuracy disparity on balanced data like ImageNet. We experimentally demonstrate that classes are not equal and the fairness issue is prevalent for image classification models across various datasets, network architectures, and model capacities. Moreover, several intriguing properties of fairness are identified. First, the unfairness lies in problematic representation rather than classifier bias. Second, with the proposed concept of Model Prediction Bias, we investigate the origins of problematic representation during optimization. Our findings reveal that models tend to exhibit greater prediction biases for classes that are more challenging to recognize. It means that more other classes will be confused with harder classes. Then the False Positives (FPs) will dominate the learning in optimization, thus leading to their poor accuracy. Further, we conclude that data augmentation and representation learning algorithms improve overall performance by promoting fairness to some degree in image classification. The Code is available at https://github.com/dvlab-research/Parametric-Contrastive-Learning.
Addressing the large distribution gap between training and testing data has long been a challenge in machine learning, giving rise to fields such as transfer learning and domain adaptation. Recently, Continuous Domain Adaptation (CDA) has emerged as an effective technique, closing this gap by utilizing a series of intermediate domains. This paper contributes a novel CDA method, W-MPOT, which rigorously addresses the domain ordering and error accumulation problems overlooked by previous studies. Specifically, we construct a transfer curriculum over the source and intermediate domains based on Wasserstein distance, motivated by theoretical analysis of CDA. Then we transfer the source model to the target domain through multiple valid paths in the curriculum using a modified version of continuous optimal transport. A bidirectional path consistency constraint is introduced to mitigate the impact of accumulated mapping errors during continuous transfer. We extensively evaluate W-MPOT on multiple datasets, achieving up to 54.1\% accuracy improvement on multi-session Alzheimer MR image classification and 94.7\% MSE reduction on battery capacity estimation.
In neural network binarization, BinaryConnect (BC) and its variants are considered the standard. These methods apply the sign function in their forward pass and their respective gradients are backpropagated to update the weights. However, the derivative of the sign function is zero whenever defined, which consequently freezes training. Therefore, implementations of BC (e.g., BNN) usually replace the derivative of sign in the backward computation with identity or other approximate gradient alternatives. Although such practice works well empirically, it is largely a heuristic or ''training trick.'' We aim at shedding some light on these training tricks from the optimization perspective. Building from existing theory on ProxConnect (PC, a generalization of BC), we (1) equip PC with different forward-backward quantizers and obtain ProxConnect++ (PC++) that includes existing binarization techniques as special cases; (2) derive a principled way to synthesize forward-backward quantizers with automatic theoretical guarantees; (3) illustrate our theory by proposing an enhanced binarization algorithm BNN++; (4) conduct image classification experiments on CNNs and vision transformers, and empirically verify that BNN++ generally achieves competitive results on binarizing these models.
Deep neural networks are typically trained using global error signals that backpropagate (BP) end-to-end, which is not only biologically implausible but also suffers from the update locking problem and requires huge memory consumption. Local learning, which updates each layer independently with a gradient-isolated auxiliary network, offers a promising alternative to address the above problems. However, existing local learning methods are confronted with a large accuracy gap with the BP counterpart, particularly for large-scale networks. This is due to the weak coupling between local layers and their subsequent network layers, as there is no gradient communication across layers. To tackle this issue, we put forward an augmented local learning method, dubbed AugLocal. AugLocal constructs each hidden layer's auxiliary network by uniformly selecting a small subset of layers from its subsequent network layers to enhance their synergy. We also propose to linearly reduce the depth of auxiliary networks as the hidden layer goes deeper, ensuring sufficient network capacity while reducing the computational cost of auxiliary networks. Our extensive experiments on four image classification datasets (i.e., CIFAR-10, SVHN, STL-10, and ImageNet) demonstrate that AugLocal can effectively scale up to tens of local layers with a comparable accuracy to BP-trained networks while reducing GPU memory usage by around 40%. The proposed AugLocal method, therefore, opens up a myriad of opportunities for training high-performance deep neural networks on resource-constrained platforms.Code is available at https://github.com/ChenxiangMA/AugLocal.
Convolutional neural networks (CNNs) have recently become the state-of-the-art tool for large-scale image classification. In this work we propose the use of activation features from CNNs as local descriptors for writer identification. A global descriptor is then formed by means of GMM supervector encoding, which is further improved by normalization with the KL-Kernel. We evaluate our method on two publicly available datasets: the ICDAR 2013 benchmark database and the CVL dataset. While we perform comparably to the state of the art on CVL, our proposed method yields about 0.21 absolute improvement in terms of mAP on the challenging bilingual ICDAR dataset.