Backpropagation-based supervised learning has achieved great success in computer vision tasks. However, its biological plausibility is always controversial. Recently, the bio-inspired Hebbian learning rule (HLR) has received extensive attention. Self-Organizing Map (SOM) uses the competitive HLR to establish connections between neurons, obtaining visual features in an unsupervised way. Although the representation of SOM neurons shows some brain-like characteristics, it is still quite different from the neuron representation in the human visual cortex. This paper proposes an improved SOM with multi-winner, multi-code, and local receptive field, named mlSOM. We observe that the neuron representation of mlSOM is similar to the human visual cortex. Furthermore, mlSOM shows a sparse distributed representation of objects, which has also been found in the human inferior temporal area. In addition, experiments show that mlSOM achieves better classification accuracy than the original SOM and other state-of-the-art HLR-based methods. The code is accessible at https://github.com/JiaHongZ/mlSOM.
Knowledge distillation (KD) has been extensively studied in single-label image classification. However, its efficacy for multi-label classification remains relatively unexplored. In this study, we firstly investigate the effectiveness of classical KD techniques, including logit-based and feature-based methods, for multi-label classification. Our findings indicate that the logit-based method is not well-suited for multi-label classification, as the teacher fails to provide inter-category similarity information or regularization effect on student model's training. Moreover, we observe that feature-based methods struggle to convey compact information of multiple labels simultaneously. Given these limitations, we propose that a suitable dark knowledge should incorporate class-wise information and be highly correlated with the final classification results. To address these issues, we introduce a novel distillation method based on Class Activation Maps (CAMs), which is both effective and straightforward to implement. Across a wide range of settings, CAMs-based distillation consistently outperforms other methods.
Deep supervision, which involves extra supervisions to the intermediate features of a neural network, was widely used in image classification in the early deep learning era since it significantly reduces the training difficulty and eases the optimization like avoiding gradient vanish over the vanilla training. Nevertheless, with the emergence of normalization techniques and residual connection, deep supervision in image classification was gradually phased out. In this paper, we revisit deep supervision for masked image modeling (MIM) that pre-trains a Vision Transformer (ViT) via a mask-and-predict scheme. Experimentally, we find that deep supervision drives the shallower layers to learn more meaningful representations, accelerates model convergence, and expands attention diversities. Our approach, called DeepMIM, significantly boosts the representation capability of each layer. In addition, DeepMIM is compatible with many MIM models across a range of reconstruction targets. For instance, using ViT-B, DeepMIM on MAE achieves 84.2 top-1 accuracy on ImageNet, outperforming MAE by +0.6. By combining DeepMIM with a stronger tokenizer CLIP, our model achieves state-of-the-art performance on various downstream tasks, including image classification (85.6 top-1 accuracy on ImageNet-1K, outperforming MAE-CLIP by +0.8), object detection (52.8 APbox on COCO) and semantic segmentation (53.1 mIoU on ADE20K). Code and models are available at https://github.com/OliverRensu/DeepMIM.
As the core building block of vision transformers, attention is a powerful tool to capture long-range dependency. However, such power comes at a cost: it incurs a huge computation burden and heavy memory footprint as pairwise token interaction across all spatial locations is computed. A series of works attempt to alleviate this problem by introducing handcrafted and content-agnostic sparsity into attention, such as restricting the attention operation to be inside local windows, axial stripes, or dilated windows. In contrast to these approaches, we propose a novel dynamic sparse attention via bi-level routing to enable a more flexible allocation of computations with content awareness. Specifically, for a query, irrelevant key-value pairs are first filtered out at a coarse region level, and then fine-grained token-to-token attention is applied in the union of remaining candidate regions (\ie, routed regions). We provide a simple yet effective implementation of the proposed bi-level routing attention, which utilizes the sparsity to save both computation and memory while involving only GPU-friendly dense matrix multiplications. Built with the proposed bi-level routing attention, a new general vision transformer, named BiFormer, is then presented. As BiFormer attends to a small subset of relevant tokens in a \textbf{query adaptive} manner without distraction from other irrelevant ones, it enjoys both good performance and high computational efficiency, especially in dense prediction tasks. Empirical results across several computer vision tasks such as image classification, object detection, and semantic segmentation verify the effectiveness of our design. Code is available at \url{https://github.com/rayleizhu/BiFormer}.
The quadratic computational complexity to the number of tokens limits the practical applications of Vision Transformers (ViTs). Several works propose to prune redundant tokens to achieve efficient ViTs. However, these methods generally suffer from (i) dramatic accuracy drops, (ii) application difficulty in the local vision transformer, and (iii) non-general-purpose networks for downstream tasks. In this work, we propose a novel Semantic Token ViT (STViT), for efficient global and local vision transformers, which can also be revised to serve as backbone for downstream tasks. The semantic tokens represent cluster centers, and they are initialized by pooling image tokens in space and recovered by attention, which can adaptively represent global or local semantic information. Due to the cluster properties, a few semantic tokens can attain the same effect as vast image tokens, for both global and local vision transformers. For instance, only 16 semantic tokens on DeiT-(Tiny,Small,Base) can achieve the same accuracy with more than 100% inference speed improvement and nearly 60% FLOPs reduction; on Swin-(Tiny,Small,Base), we can employ 16 semantic tokens in each window to further speed it up by around 20% with slight accuracy increase. Besides great success in image classification, we also extend our method to video recognition. In addition, we design a STViT-R(ecover) network to restore the detailed spatial information based on the STViT, making it work for downstream tasks, which is powerless for previous token sparsification methods. Experiments demonstrate that our method can achieve competitive results compared to the original networks in object detection and instance segmentation, with over 30% FLOPs reduction for backbone. Code is available at http://github.com/changsn/STViT-R
While Multiple Instance Learning (MIL) has shown promising results in digital Pathology Whole Slide Image (WSI) classification, such a paradigm still faces performance and generalization problems due to challenges in high computational costs on Gigapixel WSIs and limited sample size for model training. To deal with the computation problem, most MIL methods utilize a frozen pretrained model from ImageNet to obtain representations first. This process may lose essential information owing to the large domain gap and hinder the generalization of model due to the lack of image-level training-time augmentations. Though Self-supervised Learning (SSL) proposes viable representation learning schemes, the improvement of the downstream task still needs to be further explored in the conversion from the task-agnostic features of SSL to the task-specifics under the partial label supervised learning. To alleviate the dilemma of computation cost and performance, we propose an efficient WSI fine-tuning framework motivated by the Information Bottleneck theory. The theory enables the framework to find the minimal sufficient statistics of WSI, thus supporting us to fine-tune the backbone into a task-specific representation only depending on WSI-level weak labels. The WSI-MIL problem is further analyzed to theoretically deduce our fine-tuning method. Our framework is evaluated on five pathology WSI datasets on various WSI heads. The experimental results of our fine-tuned representations show significant improvements in both accuracy and generalization compared with previous works. Source code will be available at https://github.com/invoker-LL/WSI-finetuning.
Knowledge Transfer (KT) achieves competitive performance and is widely used for image classification tasks in model compression and transfer learning. Existing KT works transfer the information from a large model ("teacher") to train a small model ("student") by minimizing the difference of their conditionally independent output distributions. However, these works overlook the high-dimension structural knowledge from the intermediate representations of the teacher, which leads to limited effectiveness, and they are motivated by various heuristic intuitions, which makes it difficult to generalize. This paper proposes a novel Contrastive Knowledge Transfer Framework (CKTF), which enables the transfer of sufficient structural knowledge from the teacher to the student by optimizing multiple contrastive objectives across the intermediate representations between them. Also, CKTF provides a generalized agreement to existing KT techniques and increases their performance significantly by deriving them as specific cases of CKTF. The extensive evaluation shows that CKTF consistently outperforms the existing KT works by 0.04% to 11.59% in model compression and by 0.4% to 4.75% in transfer learning on various models and datasets.
Model compression is a crucial part of deploying neural networks (NNs), especially when the memory and storage of computing devices are limited in many applications. This paper focuses on two model compression techniques: low-rank approximation and weight pruning in neural networks, which are very popular nowadays. However, training NN with low-rank approximation and weight pruning always suffers significant accuracy loss and convergence issues. In this paper, a holistic framework is proposed for model compression from a novel perspective of nonconvex optimization by designing an appropriate objective function. Then, we introduce NN-BCD, a block coordinate descent (BCD) algorithm to solve the nonconvex optimization. One advantage of our algorithm is that an efficient iteration scheme can be derived with closed-form, which is gradient-free. Therefore, our algorithm will not suffer from vanishing/exploding gradient problems. Furthermore, with the Kurdyka-{\L}ojasiewicz (K{\L}) property of our objective function, we show that our algorithm globally converges to a critical point at the rate of O(1/k), where k denotes the number of iterations. Lastly, extensive experiments with tensor train decomposition and weight pruning demonstrate the efficiency and superior performance of the proposed framework. Our code implementation is available at https://github.com/ChenyangLi-97/NN-BCD
Since large number of high-quality remote sensing images are readily accessible, exploiting the corpus of images with less manual annotation draws increasing attention. Self-supervised models acquire general feature representations by formulating a pretext task that generates pseudo-labels for massive unlabeled data to provide supervision for training. While prior studies have explored multiple self-supervised learning techniques in remote sensing domain, pretext tasks based on local-global view alignment remain underexplored, despite achieving state-of-the-art results on natural imagery. Inspired by DINO, which employs an effective representation learning structure with knowledge distillation based on global-local view alignment, we formulate two pretext tasks for self-supervised learning on remote sensing imagery (SSLRS). Using these tasks, we explore the effectiveness of positive temporal contrast as well as multi-sized views on SSLRS. We extend DINO and propose DINO-MC which uses local views of various sized crops instead of a single fixed size in order to alleviate the limited variation in object size observed in remote sensing imagery. Our experiments demonstrate that even when pre-trained on only 10% of the dataset, DINO-MC performs on par or better than existing state-of-the-art SSLRS methods on multiple remote sensing tasks, while using less computational resources. All codes, models, and results are released at https://github.com/WennyXY/DINO-MC.
Foundation models trained on large-scale dataset gain a recent surge in CV and NLP. In contrast, development in biomedical domain lags far behind due to data scarcity. To address this issue, we build and release PMC-OA, a biomedical dataset with 1.6M image-caption pairs collected from PubMedCentral's OpenAccess subset, which is 8 times larger than before. PMC-OA covers diverse modalities or diseases, with majority of the image-caption samples aligned at finer-grained level, i.e., subfigure and subcaption. While pretraining a CLIP-style model on PMC-OA, our model named PMC-CLIP achieves state-of-the-art results on various downstream tasks, including image-text retrieval on ROCO, MedMNIST image classification, Medical VQA, i.e. +8.1% R@10 on image-text retrieval, +3.9% accuracy on image classification.
Neural Architecture Search (NAS) has been increasingly appealing to the society of object Re-Identification (ReID), for that task-specific architectures significantly improve the retrieval performance. Previous works explore new optimizing targets and search spaces for NAS ReID, yet they neglect the difference of training schemes between image classification and ReID. In this work, we propose a novel Twins Contrastive Mechanism (TCM) to provide more appropriate supervision for ReID architecture search. TCM reduces the category overlaps between the training and validation data, and assists NAS in simulating real-world ReID training schemes. We then design a Multi-Scale Interaction (MSI) search space to search for rational interaction operations between multi-scale features. In addition, we introduce a Spatial Alignment Module (SAM) to further enhance the attention consistency confronted with images from different sources. Under the proposed NAS scheme, a specific architecture is automatically searched, named as MSINet. Extensive experiments demonstrate that our method surpasses state-of-the-art ReID methods on both in-domain and cross-domain scenarios. Source code available in https://github.com/vimar-gu/MSINet.
In recent decades, wildfires, as widespread and extremely destructive natural disasters, have caused tremendous property losses and fatalities, as well as extensive damage to forest ecosystems. Many fire risk assessment projects have been proposed to prevent wildfires, but GIS-based methods are inherently challenging to scale to different geographic areas due to variations in data collection and local conditions. Inspired by the abundance of publicly available remote sensing projects and the burgeoning development of deep learning in computer vision, our research focuses on assessing fire risk using remote sensing imagery. In this work, we propose a novel remote sensing dataset, FireRisk, consisting of 7 fire risk classes with a total of 91872 labelled images for fire risk assessment. This remote sensing dataset is labelled with the fire risk classes supplied by the Wildfire Hazard Potential (WHP) raster dataset, and remote sensing images are collected using the National Agriculture Imagery Program (NAIP), a high-resolution remote sensing imagery program. On FireRisk, we present benchmark performance for supervised and self-supervised representations, with Masked Autoencoders (MAE) pre-trained on ImageNet1k achieving the highest classification accuracy, 65.29%. This remote sensing dataset, FireRisk, provides a new direction for fire risk assessment, and we make it publicly available on https://github.com/CharmonyShen/FireRisk.
Convolutional Neural Networks (CNNs) have reigned for a decade as the de facto approach to automated medical image diagnosis, pushing the state-of-the-art in classification, detection and segmentation tasks. Over the last years, vision transformers (ViTs) have appeared as a competitive alternative to CNNs, yielding impressive levels of performance in the natural image domain, while possessing several interesting properties that could prove beneficial for medical imaging tasks. In this work, we explore the benefits and drawbacks of transformer-based models for medical image classification. We conduct a series of experiments on several standard 2D medical image benchmark datasets and tasks. Our findings show that, while CNNs perform better if trained from scratch, off-the-shelf vision transformers can perform on par with CNNs when pretrained on ImageNet, both in a supervised and self-supervised setting, rendering them as a viable alternative to CNNs.
While features of different scales are perceptually important to visual inputs, existing vision transformers do not yet take advantage of them explicitly. To this end, we first propose a cross-scale vision transformer, CrossFormer. It introduces a cross-scale embedding layer (CEL) and a long-short distance attention (LSDA). On the one hand, CEL blends each token with multiple patches of different scales, providing the self-attention module itself with cross-scale features. On the other hand, LSDA splits the self-attention module into a short-distance one and a long-distance counterpart, which not only reduces the computational burden but also keeps both small-scale and large-scale features in the tokens. Moreover, through experiments on CrossFormer, we observe another two issues that affect vision transformers' performance, i.e., the enlarging self-attention maps and amplitude explosion. Thus, we further propose a progressive group size (PGS) paradigm and an amplitude cooling layer (ACL) to alleviate the two issues, respectively. The CrossFormer incorporating with PGS and ACL is called CrossFormer++. Extensive experiments show that CrossFormer++ outperforms the other vision transformers on image classification, object detection, instance segmentation, and semantic segmentation tasks. The code will be available at: https://github.com/cheerss/CrossFormer.
Learning from noisy data is a challenging task that significantly degenerates the model performance. In this paper, we present TCL, a novel twin contrastive learning model to learn robust representations and handle noisy labels for classification. Specifically, we construct a Gaussian mixture model (GMM) over the representations by injecting the supervised model predictions into GMM to link label-free latent variables in GMM with label-noisy annotations. Then, TCL detects the examples with wrong labels as the out-of-distribution examples by another two-component GMM, taking into account the data distribution. We further propose a cross-supervision with an entropy regularization loss that bootstraps the true targets from model predictions to handle the noisy labels. As a result, TCL can learn discriminative representations aligned with estimated labels through mixup and contrastive learning. Extensive experimental results on several standard benchmarks and real-world datasets demonstrate the superior performance of TCL. In particular, TCL achieves 7.5\% improvements on CIFAR-10 with 90\% noisy label -- an extremely noisy scenario. The source code is available at \url{https://github.com/Hzzone/TCL}.
In this paper, we study a novel inference paradigm, termed as schema inference, that learns to deductively infer the explainable predictions by rebuilding the prior deep neural network (DNN) forwarding scheme, guided by the prevalent philosophical cognitive concept of schema. We strive to reformulate the conventional model inference pipeline into a graph matching policy that associates the extracted visual concepts of an image with the pre-computed scene impression, by analogy with human reasoning mechanism via impression matching. To this end, we devise an elaborated architecture, termed as SchemaNet, as a dedicated instantiation of the proposed schema inference concept, that models both the visual semantics of input instances and the learned abstract imaginations of target categories as topological relational graphs. Meanwhile, to capture and leverage the compositional contributions of visual semantics in a global view, we also introduce a universal Feat2Graph scheme in SchemaNet to establish the relational graphs that contain abundant interaction information. Both the theoretical analysis and the experimental results on several benchmarks demonstrate that the proposed schema inference achieves encouraging performance and meanwhile yields a clear picture of the deductive process leading to the predictions. Our code is available at https://github.com/zhfeing/SchemaNet-PyTorch.
Training stability is of great importance to Transformers. In this work, we investigate the training dynamics of Transformers by examining the evolution of the attention layers. In particular, we track the attention entropy for each attention head during the course of training, which is a proxy for model sharpness. We identify a common pattern across different architectures and tasks, where low attention entropy is accompanied by high training instability, which can take the form of oscillating loss or divergence. We denote the pathologically low attention entropy, corresponding to highly concentrated attention scores, as $\textit{entropy collapse}$. As a remedy, we propose $\sigma$Reparam, a simple and efficient solution where we reparametrize all linear layers with spectral normalization and an additional learned scalar. We demonstrate that $\sigma$Reparam successfully prevents entropy collapse in the attention layers, promoting more stable training. Additionally, we prove a tight lower bound of the attention entropy, which decreases exponentially fast with the spectral norm of the attention logits, providing additional motivation for our approach. We conduct experiments with $\sigma$Reparam on image classification, image self-supervised learning, machine translation, speech recognition, and language modeling tasks. We show that $\sigma$Reparam provides stability and robustness with respect to the choice of hyperparameters, going so far as enabling training (a) a Vision Transformer {to competitive performance} without warmup, weight decay, layer normalization or adaptive optimizers; (b) deep architectures in machine translation and (c) speech recognition to competitive performance without warmup and adaptive optimizers. Code is available at \url{https://github.com/apple/ml-sigma-reparam}.
Modern machine learning models may be susceptible to learning spurious correlations that hold on average but not for the atypical group of samples. To address the problem, previous approaches minimize the empirical worst-group risk. Despite the promise, they often assume that each sample belongs to one and only one group, which does not allow expressing the uncertainty in group labeling. In this paper, we propose a novel framework PG-DRO, which explores the idea of probabilistic group membership for distributionally robust optimization. Key to our framework, we consider soft group membership instead of hard group annotations. The group probabilities can be flexibly generated using either supervised learning or zero-shot approaches. Our framework accommodates samples with group membership ambiguity, offering stronger flexibility and generality than the prior art. We comprehensively evaluate PG-DRO on both image classification and natural language processing benchmarks, establishing superior performance
Transferring knowledge from a source domain to a target domain can be crucial for whole slide image classification, since the number of samples in a dataset is often limited due to high annotation costs. However, domain shift and task discrepancy between datasets can hinder effective knowledge transfer. In this paper, we propose a Target-Aware Knowledge Transfer framework, employing a teacher-student paradigm. Our framework enables the teacher model to learn common knowledge from the source and target domains by actively incorporating unlabelled target images into the training of the teacher model. The teacher bag features are subsequently adapted to supervise the training of the student model on the target domain. Despite incorporating the target features during training, the teacher model tends to overlook them under the inherent domain shift and task discrepancy. To alleviate this, we introduce a target-aware feature alignment module to establish a transferable latent relationship between the source and target features by solving the optimal transport problem. Experimental results show that models employing knowledge transfer outperform those trained from scratch, and our method achieves state-of-the-art performance among other knowledge transfer methods on various datasets, including TCGA-RCC, TCGA-NSCLC, and Camelyon16.
The problem of spurious correlations (SCs) arises when a classifier relies on non-predictive features that happen to be correlated with the labels in the training data. For example, a classifier may misclassify dog breeds based on the background of dog images. This happens when the backgrounds are correlated with other breeds in the training data, leading to misclassifications during test time. Previous SC benchmark datasets suffer from varying issues, e.g., over-saturation or only containing one-to-one (O2O) SCs, but no many-to-many (M2M) SCs arising between groups of spurious attributes and classes. In this paper, we present \benchmark-\{O2O, M2M\}-\{Easy, Medium, Hard\}, an image classification benchmark suite containing spurious correlations between classes and backgrounds. To create this dataset, we employ a text-to-image model to generate photo-realistic images and an image captioning model to filter out unsuitable ones. The resulting dataset is of high quality and contains approximately 152k images. Our experimental results demonstrate that state-of-the-art group robustness methods struggle with \benchmark, most notably on the Hard-splits with none of them getting over $70\%$ accuracy on the hardest split using a ResNet50 pretrained on ImageNet. By examining model misclassifications, we detect reliances on spurious backgrounds, demonstrating that our dataset provides a significant challenge.