We introduce ProtoPool, an interpretable image classification model with a pool of prototypes shared by the classes. The training is more straightforward than in the existing methods because it does not require the pruning stage. It is obtained by introducing a fully differentiable assignment of prototypes to particular classes. Moreover, we introduce a novel focal similarity function to focus the model on the rare foreground features. We show that ProtoPool obtains state-of-the-art accuracy on the CUB-200-2011 and the Stanford Cars datasets, substantially reducing the number of prototypes. We provide a theoretical analysis of the method and a user study to show that our prototypes are more distinctive than those obtained with competitive methods.
Model compression can significantly reduce the sizes of deep neural network (DNN) models, and thus facilitates the dissemination of sophisticated, sizable DNN models, especially for their deployment on mobile or embedded devices. However, the prediction results of compressed models may deviate from those of their original models. To help developers thoroughly understand the impact of model compression, it is essential to test these models to find those deviated behaviors before dissemination. However, this is a non-trivial task because the architectures and gradients of compressed models are usually not available. To this end, we propose DFLARE, a novel, search-based, black-box testing technique to automatically find triggering inputs that result in deviated behaviors in image classification tasks. DFLARE iteratively applies a series of mutation operations to a given seed image, until a triggering input is found. For better efficacy and efficiency, DFLARE models the search problem as Markov Chains and leverages the Metropolis-Hasting algorithm to guide the selection of mutation operators in each iteration. Further, DFLARE utilizes a novel fitness function to prioritize the mutated inputs that either cause large differences between two models' outputs, or trigger previously unobserved models' probability vectors. We evaluated DFLARE on 21 compressed models for image classification tasks with three datasets. The results show that DFLARE outperforms the baseline in terms of efficacy and efficiency. We also demonstrated that the triggering inputs found by DFLARE can be used to repair up to 48.48% deviated behaviors in image classification tasks and further decrease the effectiveness of DFLARE on the repaired models.
Object detection has achieved a huge breakthrough with deep neural networks and massive annotated data. However, current detection methods cannot be directly transferred to the scenario where the annotated data is scarce due to the severe overfitting problem. Although few-shot learning and zero-shot learning have been extensively explored in the field of image classification, it is indispensable to design new methods for object detection in the data-scarce scenario since object detection has an additional challenging localization task. Low-Shot Object Detection (LSOD) is an emerging research topic of detecting objects from a few or even no annotated samples, consisting of One-Shot Object Detection (OSOD), Few-Shot Object Detection (FSOD) and Zero-Shot Object Detection (ZSD). This survey provides a comprehensive review of LSOD methods. First, we propose a thorough taxonomy of LSOD methods and analyze them systematically, comprising some extensional topics of LSOD (semi-supervised LSOD, weakly-supervised LSOD, and incremental LSOD). Then, we indicate the pros and cons of current LSOD methods with a comparison of their performance. Finally, we discuss the challenges and promising directions of LSOD to provide guidance for future works.
We introduce a new setting of Novel Class Discovery in Semantic Segmentation (NCDSS), which aims at segmenting unlabeled images containing new classes given prior knowledge from a labeled set of disjoint classes. In contrast to existing approaches that look at novel class discovery in image classification, we focus on the more challenging semantic segmentation. In NCDSS, we need to distinguish the objects and background, and to handle the existence of multiple classes within an image, which increases the difficulty in using the unlabeled data. To tackle this new setting, we leverage the labeled base data and a saliency model to coarsely cluster novel classes for model training in our basic framework. Additionally, we propose the Entropy-based Uncertainty Modeling and Self-training (EUMS) framework to overcome noisy pseudo-labels, further improving the model performance on the novel classes. Our EUMS utilizes an entropy ranking technique and a dynamic reassignment to distill clean labels, thereby making full use of the noisy data via self-supervised learning. We build the NCDSS benchmark on the PASCAL-5$^i$ dataset and COCO-20$^i$ dataset. Extensive experiments demonstrate the feasibility of the basic framework (achieving an average mIoU of 49.81% on PASCAL-5$^i$) and the effectiveness of EUMS framework (outperforming the basic framework by 9.28% mIoU on PASCAL-5$^i$).
The vision transformer splits each image into a sequence of tokens with fixed length and processes the tokens in the same way as words in natural language processing. More tokens normally lead to better performance but considerably increased computational cost. Motivated by the proverb "A picture is worth a thousand words" we aim to accelerate the ViT model by making a long image short. To this end, we propose a novel approach to assign token length adaptively during inference. Specifically, we first train a ViT model, called Resizable-ViT (ReViT), that can process any given input with diverse token lengths. Then, we retrieve the "token-length label" from ReViT and use it to train a lightweight Token-Length Assigner (TLA). The token-length labels are the smallest number of tokens to split an image that the ReViT can make the correct prediction, and TLA is learned to allocate the optimal token length based on these labels. The TLA enables the ReViT to process the image with the minimum sufficient number of tokens during inference. Thus, the inference speed is boosted by reducing the token numbers in the ViT model. Our approach is general and compatible with modern vision transformer architectures and can significantly reduce computational expanse. We verified the effectiveness of our methods on multiple representative ViT models (DeiT, LV-ViT, and TimesFormer) across two tasks (image classification and action recognition).
In many applications of computer vision it is important to accurately estimate the trajectory of an object over time by fusing data from a number of sources, of which 2D and 3D imagery is only one. In this paper, we show how to use a deep feature encoding in conjunction with generative densities over the features in a factor-graph based, probabilistic tracking framework. We present a likelihood model that combines a learned feature encoder with generative densities over them, both trained in a supervised manner. We also experiment with directly inferring probability through the use of image classification models that feed into the likelihood formulation. These models are used to implement deep factors that are added to the factor graph to complement other factors that represent domain-specific knowledge such as motion models and/or other prior information. Factors are then optimized together in a non-linear least-squares tracking framework that takes the form of an Extended Kalman Smoother with a Gaussian prior. A key feature of our likelihood model is that it leverages the Lie group properties of the tracked target's pose to apply the feature encoding on an image patch, extracted through a differentiable warp function inspired by spatial transformer networks. To illustrate the proposed approach we evaluate it on a challenging social insect behavior dataset, and show that using deep features does outperform these earlier linear appearance models used in this setting.
Recently, RobustBench (Croce et al. 2020) has become a widely recognized benchmark for the adversarial robustness of image classification networks. In its most commonly reported sub-task, RobustBench evaluates and ranks the adversarial robustness of trained neural networks on CIFAR10 under AutoAttack (Croce and Hein 2020b) with l-inf perturbations limited to eps = 8/255. With leading scores of the currently best performing models of around 60% of the baseline, it is fair to characterize this benchmark to be quite challenging. Despite its general acceptance in recent literature, we aim to foster discussion about the suitability of RobustBench as a key indicator for robustness which could be generalized to practical applications. Our line of argumentation against this is two-fold and supported by excessive experiments presented in this paper: We argue that I) the alternation of data by AutoAttack with l-inf, eps = 8/255 is unrealistically strong, resulting in close to perfect detection rates of adversarial samples even by simple detection algorithms and human observers. We also show that other attack methods are much harder to detect while achieving similar success rates. II) That results on low-resolution data sets like CIFAR10 do not generalize well to higher resolution images as gradient-based attacks appear to become even more detectable with increasing resolutions.
With the rapid development of deep learning technology and improvement in computing capability, deep learning has been widely used in the field of hyperspectral image (HSI) classification. In general, deep learning models often contain many trainable parameters and require a massive number of labeled samples to achieve optimal performance. However, in regard to HSI classification, a large number of labeled samples is generally difficult to acquire due to the difficulty and time-consuming nature of manual labeling. Therefore, many research works focus on building a deep learning model for HSI classification with few labeled samples. In this article, we concentrate on this topic and provide a systematic review of the relevant literature. Specifically, the contributions of this paper are twofold. First, the research progress of related methods is categorized according to the learning paradigm, including transfer learning, active learning and few-shot learning. Second, a number of experiments with various state-of-the-art approaches has been carried out, and the results are summarized to reveal the potential research directions. More importantly, it is notable that although there is a vast gap between deep learning models (that usually need sufficient labeled samples) and the HSI scenario with few labeled samples, the issues of small-sample sets can be well characterized by fusion of deep learning methods and related techniques, such as transfer learning and a lightweight model. For reproducibility, the source codes of the methods assessed in the paper can be found at https://github.com/ShuGuoJ/HSI-Classification.git.
Imperfect labels are ubiquitous in real-world datasets and seriously harm the model performance. Several recent effective methods for handling noisy labels have two key steps: 1) dividing samples into cleanly labeled and wrongly labeled sets by training loss, 2) using semi-supervised methods to generate pseudo-labels for samples in the wrongly labeled set. However, current methods always hurt the informative hard samples due to the similar loss distribution between the hard samples and the noisy ones. In this paper, we proposed PGDF (Prior Guided Denoising Framework), a novel framework to learn a deep model to suppress noise by generating the samples' prior knowledge, which is integrated into both dividing samples step and semi-supervised step. Our framework can save more informative hard clean samples into the cleanly labeled set. Besides, our framework also promotes the quality of pseudo-labels during the semi-supervised step by suppressing the noise in the current pseudo-labels generating scheme. To further enhance the hard samples, we reweight the samples in the cleanly labeled set during training. We evaluated our method using synthetic datasets based on CIFAR-10 and CIFAR-100, as well as on the real-world datasets WebVision and Clothing1M. The results demonstrate substantial improvements over state-of-the-art methods.
Multi-label few-shot image classification (ML-FSIC) is the task of assigning descriptive labels to previously unseen images, based on a small number of training examples. A key feature of the multi-label setting is that images often have multiple labels, which typically refer to different regions of the image. When estimating prototypes, in a metric-based setting, it is thus important to determine which regions are relevant for which labels, but the limited amount of training data makes this highly challenging. As a solution, in this paper we propose to use word embeddings as a form of prior knowledge about the meaning of the labels. In particular, visual prototypes are obtained by aggregating the local feature maps of the support images, using an attention mechanism that relies on the label embeddings. As an important advantage, our model can infer prototypes for unseen labels without the need for fine-tuning any model parameters, which demonstrates its strong generalization abilities. Experiments on COCO and PASCAL VOC furthermore show that our model substantially improves the current state-of-the-art.
Transformer is a potentially powerful architecture for vision tasks. Although equipped with more parameters and attention mechanism, its performance is not as dominant as CNN currently. CNN is usually computationally cheaper and still the leading competitor in various vision tasks. One research direction is to adopt the successful ideas of CNN and improve transformer, but it often relies on elaborated and heuristic network design. Observing that transformer and CNN are complementary in representation learning and convergence speed, we propose an efficient training framework called Vision Pair Learning (VPL) for image classification task. VPL builds up a network composed of a transformer branch, a CNN branch and pair learning module. With multi-stage training strategy, VPL enables the branches to learn from their partners during the appropriate stage of the training process, and makes them both achieve better performance with less time cost. Without external data, VPL promotes the top-1 accuracy of ViT-Base and ResNet-50 on the ImageNet-1k validation set to 83.47% and 79.61% respectively. Experiments on other datasets of various domains prove the efficacy of VPL and suggest that transformer performs better when paired with the differently structured CNN in VPL. we also analyze the importance of components through ablation study.
In edge-cloud collaborative intelligence (CI), an unreliable transmission channel exists in the information path of the AI model performing the inference. It is important to be able to simulate the performance of the CI system across an imperfect channel in order to understand system behavior and develop appropriate error control strategies. In this paper we present a simulation framework called DFTS2, which enables researchers to define the components of the CI system in TensorFlow~2, select a packet-based channel model with various parameters, and simulate system behavior under various channel conditions and error/loss control strategies. Using DFTS2, we also present the most comprehensive study to date of the packet loss concealment methods for collaborative image classification models.
While Knowledge Distillation (KD) has been recognized as a useful tool in many visual tasks, such as supervised classification and self-supervised representation learning, the main drawback of a vanilla KD framework is its mechanism, which consumes the majority of the computational overhead on forwarding through the giant teacher networks, making the entire learning procedure inefficient and costly. ReLabel, a recently proposed solution, suggests creating a label map for the entire image. During training, it receives the cropped region-level label by RoI aligning on a pre-generated entire label map, allowing for efficient supervision generation without having to pass through the teachers many times. However, as the KD teachers are from conventional multi-crop training, there are various mismatches between the global label-map and region-level label in this technique, resulting in performance deterioration. In this study, we present a Fast Knowledge Distillation (FKD) framework that replicates the distillation training phase and generates soft labels using the multi-crop KD approach, while training faster than ReLabel since no post-processes such as RoI align and softmax operations are used. When conducting multi-crop in the same image for data loading, our FKD is even more efficient than the traditional image classification framework. On ImageNet-1K, we obtain 79.8% with ResNet-50, outperforming ReLabel by ~1.0% while being faster. On the self-supervised learning task, we also show that FKD has an efficiency advantage. Our project page: http://zhiqiangshen.com/projects/FKD/index.html, source code and models are available at: https://github.com/szq0214/FKD.
In this paper, we study Multiscale Vision Transformers (MViTv2) as a unified architecture for image and video classification, as well as object detection. We present an improved version of MViT that incorporates decomposed relative positional embeddings and residual pooling connections. We instantiate this architecture in five sizes and evaluate it for ImageNet classification, COCO detection and Kinetics video recognition where it outperforms prior work. We further compare MViTv2s' pooling attention to window attention mechanisms where it outperforms the latter in accuracy/compute. Without bells-and-whistles, MViTv2 has state-of-the-art performance in 3 domains: 88.8% accuracy on ImageNet classification, 58.7 boxAP on COCO object detection as well as 86.1% on Kinetics-400 video classification. Code and models are available at https://github.com/facebookresearch/mvit.
Split learning is a popular technique used for vertical federated learning (VFL), where the goal is to jointly train a model on the private input and label data held by two parties. This technique uses a split-model, trained end-to-end, by exchanging the intermediate representations (IR) of the inputs and gradients of the IR between the two parties. We propose ExPLoit - a label-leakage attack that allows an adversarial input-owner to extract the private labels of the label-owner during split-learning. ExPLoit frames the attack as a supervised learning problem by using a novel loss function that combines gradient-matching and several regularization terms developed using key properties of the dataset and models. Our evaluations show that ExPLoit can uncover the private labels with near-perfect accuracy of up to 99.96%. Our findings underscore the need for better training techniques for VFL.
In recent years, the security of AI systems has drawn increasing research attention, especially in the medical imaging realm. To develop a secure medical image analysis (MIA) system, it is a must to study possible backdoor attacks (BAs), which can embed hidden malicious behaviors into the system. However, designing a unified BA method that can be applied to various MIA systems is challenging due to the diversity of imaging modalities (e.g., X-Ray, CT, and MRI) and analysis tasks (e.g., classification, detection, and segmentation). Most existing BA methods are designed to attack natural image classification models, which apply spatial triggers to training images and inevitably corrupt the semantics of poisoned pixels, leading to the failures of attacking dense prediction models. To address this issue, we propose a novel Frequency-Injection based Backdoor Attack method (FIBA) that is capable of delivering attacks in various MIA tasks. Specifically, FIBA leverages a trigger function in the frequency domain that can inject the low-frequency information of a trigger image into the poisoned image by linearly combining the spectral amplitude of both images. Since it preserves the semantics of the poisoned image pixels, FIBA can perform attacks on both classification and dense prediction models. Experiments on three benchmarks in MIA (i.e., ISIC-2019 for skin lesion classification, KiTS-19 for kidney tumor segmentation, and EAD-2019 for endoscopic artifact detection), validate the effectiveness of FIBA and its superiority over state-of-the-art methods in attacking MIA models as well as bypassing backdoor defense. Source code will be available at https://github.com/HazardFY/FIBA.
Learning from mistakes is an effective learning approach widely used in human learning, where a learner pays greater focus on mistakes to circumvent them in the future to improve the overall learning outcomes. In this work, we aim to investigate how effectively we can leverage this exceptional learning ability to improve machine learning models. We propose a simple and effective multi-level optimization framework called learning from mistakes using class weighting (LFM-CW), inspired by mistake-driven learning to train better machine learning models. In this formulation, the primary objective is to train a model to perform effectively on target tasks by using a re-weighting technique. We learn the class weights by minimizing the validation loss of the model and re-train the model with the synthetic data from the image generator weighted by class-wise performance and real data. We apply our LFM-CW framework with differential architecture search methods on image classification datasets such as CIFAR and ImageNet, where the results show that our proposed strategy achieves lower error rate than the baselines.
The recent success of the generative model shows that leveraging the multi-modal embedding space can manipulate an image using text information. However, manipulating an image with other sources rather than text, such as sound, is not easy due to the dynamic characteristics of the sources. Especially, sound can convey vivid emotions and dynamic expressions of the real world. Here, we propose a framework that directly encodes sound into the multi-modal (image-text) embedding space and manipulates an image from the space. Our audio encoder is trained to produce a latent representation from an audio input, which is forced to be aligned with image and text representations in the multi-modal embedding space. We use a direct latent optimization method based on aligned embeddings for sound-guided image manipulation. We also show that our method can mix text and audio modalities, which enrich the variety of the image modification. We verify the effectiveness of our sound-guided image manipulation quantitatively and qualitatively. We also show that our method can mix different modalities, i.e., text and audio, which enrich the variety of the image modification. The experiments on zero-shot audio classification and semantic-level image classification show that our proposed model outperforms other text and sound-guided state-of-the-art methods.
Due to the advent of modern embedded systems and mobile devices with constrained resources, there is a great demand for incredibly efficient deep neural networks for machine learning purposes. There is also a growing concern of privacy and confidentiality of user data within the general public when their data is processed and stored in an external server which has further fueled the need for developing such efficient neural networks for real-time inference on local embedded systems. The scope of our work presented in this paper is limited to image classification using a convolutional neural network. A Convolutional Neural Network (CNN) is a class of Deep Neural Network (DNN) widely used in the analysis of visual images captured by an image sensor, designed to extract information and convert it into meaningful representations for real-time inference of the input data. In this paper, we propose a neoteric variant of deep convolutional neural network architecture to ameliorate the performance of existing CNN architectures for real-time inference on embedded systems. We show that this architecture, dubbed CondenseNeXt, is remarkably efficient in comparison to the baseline neural network architecture, CondenseNet, by reducing trainable parameters and FLOPs required to train the network whilst maintaining a balance between the trained model size of less than 3.0 MB and accuracy trade-off resulting in an unprecedented computational efficiency.
The pretrain-finetune paradigm is a classical pipeline in visual learning. Recent progress on unsupervised pretraining methods shows superior transfer performance to their supervised counterparts. This paper revisits this phenomenon and sheds new light on understanding the transferability gap between unsupervised and supervised pretraining from a multilayer perceptron (MLP) perspective. While previous works focus on the effectiveness of MLP on unsupervised image classification where pretraining and evaluation are conducted on the same dataset, we reveal that the MLP projector is also the key factor to better transferability of unsupervised pretraining methods than supervised pretraining methods. Based on this observation, we attempt to close the transferability gap between supervised and unsupervised pretraining by adding an MLP projector before the classifier in supervised pretraining. Our analysis indicates that the MLP projector can help retain intra-class variation of visual features, decrease the feature distribution distance between pretraining and evaluation datasets, and reduce feature redundancy. Extensive experiments on public benchmarks demonstrate that the added MLP projector significantly boosts the transferability of supervised pretraining, e.g. +7.2% top-1 accuracy on the concept generalization task, +5.8% top-1 accuracy for linear evaluation on 12-domain classification tasks, and +0.8% AP on COCO object detection task, making supervised pretraining comparable or even better than unsupervised pretraining.