We introduce CVNets, a high-performance open-source library for training deep neural networks for visual recognition tasks, including classification, detection, and segmentation. CVNets supports image and video understanding tools, including data loading, data transformations, novel data sampling methods, and implementations of several standard networks with similar or better performance than previous studies. Our source code is available at: \url{https://github.com/apple/ml-cvnets}.
Group convolution has been shown to offer great computational savings in various 2D convolutional architectures for image classification. It is natural to ask: 1) if group convolution can help to alleviate the high computational cost of video classification networks; 2) what factors matter the most in 3D group convolutional networks; and 3) what are good computation/accuracy trade-offs with 3D group convolutional networks. This paper studies the effects of different design choices in 3D group convolutional networks for video classification. We empirically demonstrate that the amount of channel interactions plays an important role in the accuracy of 3D group convolutional networks. Our experiments suggest two main findings. First, it is a good practice to factorize 3D convolutions by separating channel interactions and spatiotemporal interactions as this leads to improved accuracy and lower computational cost. Second, 3D channel-separated convolutions provide a form of regularization, yielding lower training accuracy but higher test accuracy compared to 3D convolutions. These two empirical findings lead us to design an architecture -- Channel-Separated Convolutional Network (CSN) -- which is simple, efficient, yet accurate. On Sports1M, Kinetics, and Something-Something, our CSNs are comparable with or better than the state-of-the-art while being 2-3 times more efficient.
Recent advancements in low-cost ensemble learning have demonstrated improved efficiency for image classification. However, the existing low-cost ensemble methods show relatively lower accuracy compared to conventional ensemble learning. In this paper, we propose a new low-cost ensemble learning, which can simultaneously achieve high efficiency and classification performance. A CNN is transformed into a multi-branch structure without introduction of additional components, which maintains the computational complexity as that of the original single model and also enhances diversity among the branches' outputs via sufficient separation between different pathways of the branches. In addition, we propose a new strategy that applies grouped convolution in the branches with different numbers of groups in different branches, which boosts the diversity of the branches' outputs. For training, we employ knowledge distillation using the ensemble of the outputs as the teacher signal. The high diversity among the outputs enables to form a powerful teacher, enhancing the individual branch's classification performance and consequently the overall ensemble performance. Experimental results show that our method achieves state-of-the-art classification accuracy and higher uncertainty estimation performance compared to previous low-cost ensemble methods. The code is available at https://github.com/hjdw2/SEMBG.
Training end-to-end networks for classifying gigapixel size histopathological images is computationally intractable. Most approaches are patch-based and first learn local representations (patch-wise) before combining these local representations to produce image-level decisions. However, dividing large tissue structures into patches limits the context available to these networks, which may reduce their ability to learn representations from clinically relevant structures. In this paper, we introduce a novel attention-based network, the Holistic ATtention Network (HATNet) to classify breast biopsy images. We streamline the histopathological image classification pipeline and show how to learn representations from gigapixel size images end-to-end. HATNet extends the bag-of-words approach and uses self-attention to encode global information, allowing it to learn representations from clinically relevant tissue structures without any explicit supervision. It outperforms the previous best network Y-Net, which uses supervision in the form of tissue-level segmentation masks, by 8%. Importantly, our analysis reveals that HATNet learns representations from clinically relevant structures, and it matches the classification accuracy of human pathologists for this challenging test set. Our source code is available at \url{https://github.com/sacmehta/HATNet}
Learning visual features from unlabeled images has proven successful for semantic categorization, often by mapping different $views$ of the same object to the same feature to achieve recognition invariance. However, visual recognition involves not only identifying $what$ an object is but also understanding $how$ it is presented. For example, seeing a car from the side versus head-on is crucial for deciding whether to stay put or jump out of the way. While unsupervised feature learning for downstream viewpoint reasoning is important, it remains under-explored, partly due to the lack of a standardized evaluation method and benchmarks. We introduce a new dataset of adjacent image triplets obtained from a viewpoint trajectory, without any semantic or pose labels. We benchmark both semantic classification and pose estimation accuracies on the same visual feature. Additionally, we propose a viewpoint trajectory regularization loss for learning features from unlabeled image triplets. Our experiments demonstrate that this approach helps develop a visual representation that encodes object identity and organizes objects by their poses, retaining semantic classification accuracy while achieving emergent global pose awareness and better generalization to novel objects. Our dataset and code are available at http://pwang.pw/trajSSL/.
Color and structure are the two pillars that construct an image. Usually, the structure is well expressed through a rich spectrum of colors, allowing objects in an image to be recognized by neural networks. However, under extreme limitations of color space, the structure tends to vanish, and thus a neural network might fail to understand the image. Interested in exploring this interplay between color and structure, we study the scientific problem of identifying and preserving the most informative image structures while constraining the color space to just a few bits, such that the resulting image can be recognized with possibly high accuracy. To this end, we propose a color quantization network, ColorCNN, which learns to structure the images from the classification loss in an end-to-end manner. Given a color space size, ColorCNN quantizes colors in the original image by generating a color index map and an RGB color palette. Then, this color-quantized image is fed to a pre-trained task network to evaluate its performance. In our experiment, with only a 1-bit color space (i.e., two colors), the proposed network achieves 82.1% top-1 accuracy on the CIFAR10 dataset, outperforming traditional color quantization methods by a large margin. For applications, when encoded with PNG, the proposed color quantization shows superiority over other image compression methods in the extremely low bit-rate regime. The code is available at: https://github.com/hou-yz/color_distillation.
Recent advances in machine learning leverage massive datasets of unlabeled images from the web to learn general-purpose image representations for tasks from image classification to face recognition. But do unsupervised computer vision models automatically learn implicit patterns and embed social biases that could have harmful downstream effects? We develop a novel method for quantifying biased associations between representations of social concepts and attributes in images. We find that state-of-the-art unsupervised models trained on ImageNet, a popular benchmark image dataset curated from internet images, automatically learn racial, gender, and intersectional biases. We replicate 8 previously documented human biases from social psychology, from the innocuous, as with insects and flowers, to the potentially harmful, as with race and gender. Our results closely match three hypotheses about intersectional bias from social psychology. For the first time in unsupervised computer vision, we also quantify implicit human biases about weight, disabilities, and several ethnicities. When compared with statistical patterns in online image datasets, our findings suggest that machine learning models can automatically learn bias from the way people are stereotypically portrayed on the web.
Volumetric image segmentation with convolutional neural networks (CNNs) encounters several challenges, which are specific to medical images. Among these challenges are large volumes of interest, high class imbalances, and difficulties in learning shape representations. To tackle these challenges, we propose to improve over traditional CNN-based volumetric image segmentation through point-wise classification of point clouds. The sparsity of point clouds allows processing of entire image volumes, balancing highly imbalanced segmentation problems, and explicitly learning an anatomical shape. We build upon PointCNN, a neural network proposed to process point clouds, and propose here to jointly encode shape and volumetric information within the point cloud in a compact and computationally effective manner. We demonstrate how this approach can then be used to refine CNN-based segmentation, which yields significantly improved results in our experiments on the difficult task of peripheral nerve segmentation from magnetic resonance neurography images. By synthetic experiments, we further show the capability of our approach in learning an explicit anatomical shape representation.
The Vision Transformer (ViT) leverages the Transformer's encoder to capture global information by dividing images into patches and achieves superior performance across various computer vision tasks. However, the self-attention mechanism of ViT captures the global context from the outset, overlooking the inherent relationships between neighboring pixels in images or videos. Transformers mainly focus on global information while ignoring the fine-grained local details. Consequently, ViT lacks inductive bias during image or video dataset training. In contrast, convolutional neural networks (CNNs), with their reliance on local filters, possess an inherent inductive bias, making them more efficient and quicker to converge than ViT with less data. In this paper, we present a lightweight Depth-Wise Convolution module as a shortcut in ViT models, bypassing entire Transformer blocks to ensure the models capture both local and global information with minimal overhead. Additionally, we introduce two architecture variants, allowing the Depth-Wise Convolution modules to be applied to multiple Transformer blocks for parameter savings, and incorporating independent parallel Depth-Wise Convolution modules with different kernels to enhance the acquisition of local information. The proposed approach significantly boosts the performance of ViT models on image classification, object detection, and instance segmentation by a large margin, especially on small datasets, as evaluated on CIFAR-10, CIFAR-100, Tiny-ImageNet and ImageNet for image classification, and COCO for object detection and instance segmentation. The source code can be accessed at https://github.com/ZTX-100/Efficient_ViT_with_DW.
Structure determination is key to understanding protein function at a molecular level. Whilst significant advances have been made in predicting structure and function from amino acid sequence, researchers must still rely on expensive, time-consuming analytical methods to visualise detailed protein conformation. In this study, we demonstrate that it is possible to make accurate ($\geq$80%) predictions of protein class and architecture from structures determined at low ($>$3A) resolution, using a deep convolutional neural network trained on high-resolution ($\leq$3A) structures represented as 2D matrices. Thus, we provide proof of concept for high-speed, low-cost protein structure classification at low resolution, and a basis for extension to prediction of function. We investigate the impact of the input representation on classification performance, showing that side-chain information may not be necessary for fine-grained structure predictions. Finally, we confirm that high-resolution, low-resolution and NMR-determined structures inhabit a common feature space, and thus provide a theoretical foundation for boosting with single-image super-resolution.
Lung and colon cancer are serious worldwide health challenges that require early and precise identification to reduce mortality risks. However, diagnosis, which is mostly dependent on histopathologists' competence, presents difficulties and hazards when expertise is insufficient. While diagnostic methods like imaging and blood markers contribute to early detection, histopathology remains the gold standard, although time-consuming and vulnerable to inter-observer mistakes. Limited access to high-end technology further limits patients' ability to receive immediate medical care and diagnosis. Recent advances in deep learning have generated interest in its application to medical imaging analysis, specifically the use of histopathological images to diagnose lung and colon cancer. The goal of this investigation is to use and adapt existing pre-trained CNN-based models, such as Xception, DenseNet201, ResNet101, InceptionV3, DenseNet121, DenseNet169, ResNet152, and InceptionResNetV2, to enhance classification through better augmentation strategies. The results show tremendous progress, with all eight models reaching impressive accuracy ranging from 97% to 99%. Furthermore, attention visualization techniques such as GradCAM, GradCAM++, ScoreCAM, Faster Score-CAM, and LayerCAM, as well as Vanilla Saliency and SmoothGrad, are used to provide insights into the models' classification decisions, thereby improving interpretability and understanding of malignant and benign image classification.
Federated learning (FL) has emerged with increasing popularity to collaborate distributed medical institutions for training deep networks. However, despite existing FL algorithms only allow the supervised training setting, most hospitals in realistic usually cannot afford the intricate data labeling due to absence of budget or expertise. This paper studies a practical yet challenging FL problem, named \textit{Federated Semi-supervised Learning} (FSSL), which aims to learn a federated model by jointly utilizing the data from both labeled and unlabeled clients (i.e., hospitals). We present a novel approach for this problem, which improves over traditional consistency regularization mechanism with a new inter-client relation matching scheme. The proposed learning scheme explicitly connects the learning across labeled and unlabeled clients by aligning their extracted disease relationships, thereby mitigating the deficiency of task knowledge at unlabeled clients and promoting discriminative information from unlabeled samples. We validate our method on two large-scale medical image classification datasets. The effectiveness of our method has been demonstrated with the clear improvements over state-of-the-arts as well as the thorough ablation analysis on both tasks\footnote{Code will be made available at \url{https://github.com/liuquande/FedIRM}}.
Existing optical flow estimators usually employ the network architectures typically designed for image classification as the encoder to extract per-pixel features. However, due to the natural difference between the tasks, the architectures designed for image classification may be sub-optimal for flow estimation. To address this issue, we propose a neural architecture search method named FlowNAS to automatically find the better encoder architecture for flow estimation task. We first design a suitable search space including various convolutional operators and construct a weight-sharing super-network for efficiently evaluating the candidate architectures. Then, for better training the super-network, we propose Feature Alignment Distillation, which utilizes a well-trained flow estimator to guide the training of super-network. Finally, a resource-constrained evolutionary algorithm is exploited to find an optimal architecture (i.e., sub-network). Experimental results show that the discovered architecture with the weights inherited from the super-network achieves 4.67\% F1-all error on KITTI, an 8.4\% reduction of RAFT baseline, surpassing state-of-the-art handcrafted models GMA and AGFlow, while reducing the model complexity and latency. The source code and trained models will be released in https://github.com/VDIGPKU/FlowNAS.
Deep learning models develop successive representations of their input in sequential layers, the last of which maps the final representation to the output. Here we investigate the informational content of these representations by observing the ability of convolutional image classification models to autoencode the model's input using embeddings existing in various layers. We find that the deeper the layer, the less accurate that layer's representation of the input is before training. Inaccurate representation results from non-uniqueness in which various distinct inputs give approximately the same embedding. Non-unique representation is a consequence of both exact and approximate non-invertibility of transformations present in the forward pass. Learning to classify natural images leads to an increase in representation clarity for early but not late layers, which instead form abstract images. Rather than simply selecting for features present in the input necessary for classification, deep layer representations are found to transform the input so that it matches representations of the training data such that arbitrary inputs are mapped to manifolds learned during training. This work provides support for the theory that the tasks of image recognition and input generation are inseparable even for models trained exclusively to classify.
In computational pathology, whole-slide image (WSI) classification presents a formidable challenge due to its gigapixel resolution and limited fine-grained annotations. Multiple-instance learning (MIL) offers a weakly supervised solution, yet refining instance-level information from bag-level labels remains challenging. While most of the conventional MIL methods use attention scores to estimate instance importance scores (IIS) which contribute to the prediction of the slide labels, these often lead to skewed attention distributions and inaccuracies in identifying crucial instances. To address these issues, we propose a new approach inspired by cooperative game theory: employing Shapley values to assess each instance's contribution, thereby improving IIS estimation. The computation of the Shapley value is then accelerated using attention, meanwhile retaining the enhanced instance identification and prioritization. We further introduce a framework for the progressive assignment of pseudo bags based on estimated IIS, encouraging more balanced attention distributions in MIL models. Our extensive experiments on CAMELYON-16, BRACS, TCGA-LUNG, and TCGA-BRCA datasets show our method's superiority over existing state-of-the-art approaches, offering enhanced interpretability and class-wise insights. Our source code is available at https://github.com/RenaoYan/PMIL.
The Gaussian kernel and its derivatives have already been employed for Convolutional Neural Networks in several previous works. Most of these papers proposed to compute filters by linearly combining one or several bases of fixed or slightly trainable Gaussian kernels with or without their derivatives. In this article, we propose a high-level configurable layer based on anisotropic, oriented and shifted Gaussian derivative kernels which generalize notions encountered in previous related works while keeping their main advantage. The results show that the proposed layer has competitive performance compared to previous works and that it can be successfully included in common deep architectures such as VGG16 for image classification and U-net for image segmentation.
Transformers and masked language modeling are quickly being adopted and explored in computer vision as vision transformers and masked image modeling (MIM). In this work, we argue that image token masking differs from token masking in text, due to the amount and correlation of tokens in an image. In particular, to generate a challenging pretext task for MIM, we advocate a shift from random masking to informed masking. We develop and exhibit this idea in the context of distillation-based MIM, where a teacher transformer encoder generates an attention map, which we use to guide masking for the student. We thus introduce a novel masking strategy, called attention-guided masking (AttMask), and we demonstrate its effectiveness over random masking for dense distillation-based MIM as well as plain distillation-based self-supervised learning on classification tokens. We confirm that AttMask accelerates the learning process and improves the performance on a variety of downstream tasks. We provide the implementation code at https://github.com/gkakogeorgiou/attmask.
Ordinal regression bridges regression and classification by assigning objects to ordered classes. While human experts rely on discriminative patch-level features for decisions, current approaches are limited by the availability of only image-level ordinal labels, overlooking fine-grained patch-level characteristics. In this paper, we propose a Dual-level Fuzzy Learning with Patch Guidance framework, named DFPG that learns precise feature-based grading boundaries from ambiguous ordinal labels, with patch-level supervision. Specifically, we propose patch-labeling and filtering strategies to enable the model to focus on patch-level features exclusively with only image-level ordinal labels available. We further design a dual-level fuzzy learning module, which leverages fuzzy logic to quantitatively capture and handle label ambiguity from both patch-wise and channel-wise perspectives. Extensive experiments on various image ordinal regression datasets demonstrate the superiority of our proposed method, further confirming its ability in distinguishing samples from difficult-to-classify categories. The code is available at https://github.com/ZJUMAI/DFPG-ord.
Semmelhack et al. (2014) have achieved high classification accuracy in distinguishing swim bouts of zebrafish using a Support Vector Machine (SVM). Convolutional Neural Networks (CNNs) have reached superior performance in various image recognition tasks over SVMs, but these powerful networks remain a black box. Reaching better transparency helps to build trust in their classifications and makes learned features interpretable to experts. Using a recently developed technique called Deep Taylor Decomposition, we generated heatmaps to highlight input regions of high relevance for predictions. We find that our CNN makes predictions by analyzing the steadiness of the tail's trunk, which markedly differs from the manually extracted features used by Semmelhack et al. (2014). We further uncovered that the network paid attention to experimental artifacts. Removing these artifacts ensured the validity of predictions. After correction, our best CNN beats the SVM by 6.12%, achieving a classification accuracy of 96.32%. Our work thus demonstrates the utility of AI explainability for CNNs.
We propose an adaptation of the curriculum training framework, applicable to state-of-the-art meta learning techniques for few-shot classification. Curriculum-based training popularly attempts to mimic human learning by progressively increasing the training complexity to enable incremental concept learning. As the meta-learner's goal is learning how to learn from as few samples as possible, the exact number of those samples (i.e. the size of the support set) arises as a natural proxy of a given task's difficulty. We define a simple yet novel curriculum schedule that begins with a larger support size and progressively reduces it throughout training to eventually match the desired shot-size of the test setup. This proposed method boosts the learning efficiency as well as the generalization capability. Our experiments with the MAML algorithm on two few-shot image classification tasks show significant gains with the curriculum training framework. Ablation studies corroborate the independence of our proposed method from the model architecture as well as the meta-learning hyperparameters