Much of the recent progress made in image classification research can be credited to training procedure refinements, such as changes in data augmentations and optimization methods. In the literature, however, most refinements are either briefly mentioned as implementation details or only visible in source code. In this paper, we will examine a collection of such refinements and empirically evaluate their impact on the final model accuracy through ablation study. We will show that, by combining these refinements together, we are able to improve various CNN models significantly. For example, we raise ResNet-50's top-1 validation accuracy from 75.3% to 79.29% on ImageNet. We will also demonstrate that improvement on image classification accuracy leads to better transfer learning performance in other application domains such as object detection and semantic segmentation.
An accurate and timely detection of diseases and pests in rice plants can help farmers in applying timely treatment on the plants and thereby can reduce the economic losses substantially. Recent developments in deep learning based convolutional neural networks (CNN) have greatly improved the image classification accuracy. Being motivated by the success of CNNs in image classification, deep learning based approaches have been developed in this paper for detecting diseases and pests from rice plant images. The contribution of this paper is two fold: (i) State-of-the-art large scale architectures such as VGG16 and InceptionV3 have been adopted and fine tuned for detecting and recognizing rice diseases and pests. Experimental results show the effectiveness of these models with real datasets. (ii) Since large scale architectures are not suitable for mobile devices, a two-stage small CNN architecture has been proposed, and compared with the state-of-the-art memory efficient CNN architectures such as MobileNet, NasNet Mobile and SqueezeNet. Experimental results show that the proposed architecture can achieve the desired accuracy of 93.3\% with a significantly reduced model size (e.g., 99\% less size compared to that of VGG16).
In large-scale statistical learning, data collection and model fitting are moving increasingly toward peripheral devices---phones, watches, fitness trackers---away from centralized data collection. Concomitant with this rise in decentralized data are increasing challenges of maintaining privacy while allowing enough information to fit accurate, useful statistical models. This motivates local notions of privacy---most significantly, local differential privacy, which provides strong protections against sensitive data disclosures---where data is obfuscated before a statistician or learner can even observe it, providing strong protections to individuals' data. Yet local privacy as traditionally employed may prove too stringent for practical use, especially in modern high-dimensional statistical and machine learning problems. Consequently, we revisit the types of disclosures and adversaries against which we provide protections, considering adversaries with limited prior information and ensuring that with high probability, ensuring they cannot reconstruct an individual's data within useful tolerances. By reconceptualizing these protections, we allow more useful data release---large privacy parameters in local differential privacy---and we design new (minimax) optimal locally differentially private mechanisms for statistical learning problems for \emph{all} privacy levels. We thus present practicable approaches to large-scale locally private model training that were previously impossible, showing theoretically and empirically that we can fit large-scale image classification and language models with little degradation in utility.
Increasing the mini-batch size for stochastic gradient descent offers significant opportunities to reduce wall-clock training time, but there are a variety of theoretical and systems challenges that impede the widespread success of this technique. We investigate these issues, with an emphasis on time to convergence and total computational cost, through an extensive empirical analysis of network training across several architectures and problem domains, including image classification, image segmentation, and language modeling. Although it is common practice to increase the batch size in order to fully exploit available computational resources, we find a substantially more nuanced picture. Our main finding is that across a wide range of network architectures and problem domains, increasing the batch size beyond a certain point yields no decrease in wall-clock time to convergence for \emph{either} train or test loss. This batch size is usually substantially below the capacity of current systems. We show that popular training strategies for large batch size optimization begin to fail before we can populate all available compute resources, and we show that the point at which these methods break down depends more on attributes like model architecture and data complexity than it does directly on the size of the dataset.
Globally modeling and reasoning over relations between regions can be beneficial for many computer vision tasks on both images and videos. Convolutional Neural Networks (CNNs) excel at modeling local relations by convolution operations, but they are typically inefficient at capturing global relations between distant regions and require stacking multiple convolution layers. In this work, we propose a new approach for reasoning globally in which a set of features are globally aggregated over the coordinate space and then projected to an interaction space where relational reasoning can be efficiently computed. After reasoning, relation-aware features are distributed back to the original coordinate space for down-stream tasks. We further present a highly efficient instantiation of the proposed approach and introduce the Global Reasoning unit (GloRe unit) that implements the coordinate-interaction space mapping by weighted global pooling and weighted broadcasting, and the relation reasoning via graph convolution on a small graph in interaction space. The proposed GloRe unit is lightweight, end-to-end trainable and can be easily plugged into existing CNNs for a wide range of tasks. Extensive experiments show our GloRe unit can consistently boost the performance of state-of-the-art backbone architectures, including ResNet, ResNeXt, SE-Net and DPN, for both 2D and 3D CNNs, on image classification, semantic segmentation and video action recognition task.
Convolutional Neural Networks (CNNs) usually use the same activation function, such as RELU, for all convolutional layers. There are performance limitations of just using RELU. In order to achieve better classification performance, reduce training and testing times, and reduce power consumption and memory usage, a new "Compressed Multi-function CNN" is developed. Google's Inception-V4, for example, is a very deep CNN that consists of 4 Inception-A blocks, 7 Inception-B blocks, and 3 Inception-C blocks. RELU is used for all convolutional layers. A new "Compressed Multi-function Inception-V4" (CMI) that can use different activation functions is created with k Inception-A blocks, m Inception-B blocks, and n Inception-C blocks where k in {1, 2, 3, 4}, m in {1, 2, 3, 4, 5, 6, 7}, n in {1, 2, 3}, and (k+m+n)<14. For performance analysis, a dataset for classifying brain MRI images into one of the four stages of Alzheimer's disease is used to compare three CMI architectures with Inception-V4 in terms of F1-score, training and testing times (related to power consumption), and memory usage (model size). Overall, simulations show that the new CMI models can outperform both the commonly used Inception-V4 and Inception-V4 using different activation functions. In the future, other "Compressed Multi-function CNNs", such as "Compressed Multi-function ResNets and DenseNets" that have a reduced number of convolutional blocks using different activation functions, will be developed to further increase classification accuracy, reduce training and testing times, reduce computational power, and reduce memory usage (model size) for building more effective healthcare systems, such as implementing accurate and convenient disease diagnosis systems on mobile devices that have limited battery power and memory.
The success of deep learning research has catapulted deep models into production systems that our society is becoming increasingly dependent on, especially in the image and video domains. However, recent work has shown that these largely uninterpretable models exhibit glaring security vulnerabilities in the presence of an adversary. In this work, we develop a powerful untargeted adversarial attack for action recognition systems in both white-box and black-box settings. Action recognition models differ from image-classification models in that their inputs contain a temporal dimension, which we explicitly target in the attack. Drawing inspiration from image classifier attacks, we create new attacks which achieve state-of-the-art success rates on a two-stream classifier trained on the UCF-101 dataset. We find that our attacks can significantly degrade a model's performance with sparsely and imperceptibly perturbed examples. We also demonstrate the transferability of our attacks to black-box action recognition systems.
Few-shot or one-shot learning of classifiers requires a significant inductive bias towards the type of task to be learned. One way to acquire this is by meta-learning on tasks similar to the target task. In this paper, we propose UMTRA, an algorithm that performs unsupervised, model-agnostic meta-learning for classification tasks. The meta-learning step of UMTRA is performed on a flat collection of unlabeled images. While we assume that these images can be grouped into a diverse set of classes and are relevant to the target task, no explicit information about the classes or any labels are needed. UMTRA uses random sampling and augmentation to create synthetic training tasks for meta-learning phase. Labels are only needed at the final target task learning step, and they can be as little as one sample per class. On the Omniglot and Mini-Imagenet few-shot learning benchmarks, UMTRA outperforms every tested approach based on unsupervised learning of representations, while alternating for the best performance with the recent CACTUs algorithm. Compared to supervised model-agnostic meta-learning approaches, UMTRA trades off some classification accuracy for a reduction in the required labels of several orders of magnitude.
Embedded deep learning platforms have witnessed two simultaneous improvements. First, the accuracy of convolutional neural networks (CNNs) has been significantly improved through the use of automated neural-architecture search (NAS) algorithms to determine CNN structure. Second, there has been increasing interest in developing hardware accelerators for CNNs that provide improved inference performance and energy consumption compared to GPUs. Such embedded deep learning platforms differ in the amount of compute resources and memory-access bandwidth, which would affect performance and energy consumption of CNNs. It is therefore critical to consider the available hardware resources in the network architecture search. To this end, we introduce TEA-DNN, a NAS algorithm targeting multi-objective optimization of execution time, energy consumption, and classification accuracy of CNN workloads on embedded architectures. TEA-DNN leverages energy and execution time measurements on embedded hardware when exploring the Pareto-optimal curves across accuracy, execution time, and energy consumption and does not require additional effort to model the underlying hardware. We apply TEA-DNN for image classification on actual embedded platforms (NVIDIA Jetson TX2 and Intel Movidius Neural Compute Stick). We highlight the Pareto-optimal operating points that emphasize the necessity to explicitly consider hardware characteristics in the search process. To the best of our knowledge, this is the most comprehensive study of Pareto-optimal models across a range of hardware platforms using actual measurements on hardware to obtain objective values.
Computer vision is difficult, partly because the desired mathematical function connecting input and output data is often complex, fuzzy and thus hard to learn. Coarse-to-fine (C2F) learning is a promising direction, but it remains unclear how it is applied to a wide range of vision problems. This paper presents a generalized C2F framework by making two technical contributions. First, we provide a unified way of C2F propagation, in which the coarse prediction (a class vector, a detected box, a segmentation mask, etc.) is encoded into a dense (pixel-level) matrix and concatenated to the original input, so that the fine model takes the same design of the coarse model but sees additional information. Second, we present a progressive training strategy which starts with feeding the ground-truth instead of the coarse output into the fine model, and gradually increases the fraction of coarse output, so that at the end of training the fine model is ready for testing. We also relate our approach to curriculum learning by showing that data difficulty keeps increasing during the training process. We apply our framework to three vision tasks including image classification, object localization and semantic segmentation, and demonstrate consistent accuracy gain compared to the baseline training strategy.
Convolutional Neural Networks (CNNs) are commonly thought to recognise objects by learning increasingly complex representations of object shapes. Some recent studies suggest a more important role of image textures. We here put these conflicting hypotheses to a quantitative test by evaluating CNNs and human observers on images with a texture-shape cue conflict. We show that ImageNet-trained CNNs are strongly biased towards recognising textures rather than shapes, which is in stark contrast to human behavioural evidence and reveals fundamentally different classification strategies. We then demonstrate that the same standard architecture (ResNet-50) that learns a texture-based representation on ImageNet is able to learn a shape-based representation instead when trained on "Stylized-ImageNet", a stylized version of ImageNet. This provides a much better fit for human behavioural performance in our well-controlled psychophysical lab setting (nine experiments totalling 48,560 psychophysical trials across 97 observers) and comes with a number of unexpected emergent benefits such as improved object detection performance and previously unseen robustness towards a wide range of image distortions, highlighting advantages of a shape-based representation.
We introduce a light-weight, power efficient, and general purpose convolutional neural network, ESPNetv2, for modeling visual and sequential data. Our network uses group point-wise and depth-wise dilated separable convolutions to learn representations from a large effective receptive field with fewer FLOPs and parameters. The performance of our network is evaluated on four different tasks: (1) object classification, (2) semantic segmentation, (3) object detection, and (4) language modeling. Experiments on these tasks, including image classification on the ImageNet and language modeling on the PenTree bank dataset, demonstrate the superior performance of our method over the state-of-the-art methods. Our network outperforms ESPNet by 4-5% and has 2-4x fewer FLOPs on the PASCAL VOC and the Cityscapes dataset. Compared to YOLOv2 on the MS-COCO object detection, ESPNetv2 delivers 4.4% higher accuracy with 6x fewer FLOPs. Our experiments show that ESPNetv2 is much more power efficient than existing state-of-the-art efficient methods including ShuffleNets and MobileNets. Our code is open-source and available at https://github.com/sacmehta/ESPNetv2
Convolutional Neural Networks (CNN) has become more popular choice for various tasks such as computer vision, speech recognition and natural language processing. Thanks to their large computational capability and throughput, GPUs ,which are not power efficient and therefore does not suit low power systems such as mobile devices, are the most common platform for both training and inferencing tasks. Recent studies has shown that FPGAs can provide a good alternative to GPUs as a CNN accelerator, due to their re-configurable nature, low power and small latency. In order for FPGA-based accelerators outperform GPUs in inference task, both the parameters of the network and the activations must be quantized. While most works use uniform quantizers for both parameters and activations, it is not always the optimal one, and a non-uniform quantizer need to be considered. In this work we introduce a custom hardware-friendly approach to implement non-uniform quantizers. In addition, we use a single scale integer representation of both parameters and activations, for both training and inference. The combined method yields a hardware efficient non-uniform quantizer, fit for real-time applications. We have tested our method on CIFAR-10 and CIFAR-100 image classification datasets with ResNet-18 and VGG-like architectures, and saw little degradation in accuracy.
Modern deep networks generally implement a certain form of shortcut connections to alleviate optimization difficulties. However, we observe that such network topology alters the nature of deep networks. In many ways, these networks behave similarly to aggregated wide networks. We thus exploit the aggregation nature of shortcut connections at a finer architectural level and place them within wide convolutional layers. We end up with a sequentially aggregated convolutional (SeqConv) layer that combines the benefits of both wide and deep representations by aggregating features of various depths in sequence. The proposed SeqConv serves as a drop-in replacement of regular wide convolutional layers and thus could be handily integrated into any backbone network. We apply SeqConv to widely adopted backbones including ResNet and ResNeXt, and conduct experiments for image classification on public benchmark datasets. Our ResNet based network with a model size of ResNet-50 easily surpasses the performance of the 2.35$\times$ larger ResNet-152, while our ResNeXt based model sets a new state-of-the-art accuracy on ImageNet classification for networks with similar model complexity. The code and pre-trained models of our work are publicly available at https://github.com/GroupOfAlchemists/SeqConv.
Distributed data-parallel algorithms aim to accelerate the training of deep neural networks by parallelizing the computation of large mini-batch gradient updates across multiple nodes. Approaches that synchronize nodes using exact distributed averaging (e.g., via AllReduce) are sensitive to stragglers and communication delays. The PushSum gossip algorithm is robust to these issues, but only performs approximate distributed averaging. This paper studies Stochastic Gradient Push (SGP), which combines PushSum with stochastic gradient updates. We prove that SGP converges to a stationary point of smooth, non-convex objectives at the same sub-linear rate as SGD, and that all nodes achieve consensus. We empirically validate the performance of SGP on image classification (ResNet-50, ImageNet) and machine translation (Transformer, WMT'16 En-De) workloads. Our code will be made publicly available.
In this paper, we propose to train convolutional neural networks (CNNs) with both binarized weights and activations, leading to quantized models specifically} for mobile devices with limited power capacity and computation resources. Previous works on quantizing CNNs seek to approximate the floating-point information using a set of discrete values, which we call value approximation, but typically assume the same architecture as the full-precision networks. In this paper, however, we take a novel 'structure approximation' view for quantization---it is very likely that a different architecture may be better for best performance. In particular, we propose a `network decomposition' strategy, named \textbf{Group-Net}, in which we divide the network into groups. In this way, each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches. In addition, we learn effective connections among groups to improve the representational capability. Moreover, the proposed Group-Net shows strong generalization to other tasks. For instance, we extend Group-Net for highly accurate semantic segmentation by embedding rich context into the binary structure. Experiments on both classification and semantic segmentation tasks demonstrate the superior performance of the proposed methods over various popular architectures. In particular, we outperform the previous best binary neural networks in terms of accuracy and major computation savings.
Neural Architecture Search has shown potential to automate the design of neural networks. Deep Reinforcement Learning based agents can learn complex architectural patterns, as well as explore a vast and compositional search space. On the other hand, evolutionary algorithms offer higher sample efficiency, which is critical for such a resource intensive application. In order to capture the best of both worlds, we propose a class of Evolutionary-Neural hybrid agents (Evo-NAS). We show that the Evo-NAS agent outperforms both neural and evolutionary agents when applied to architecture search for a suite of text and image classification benchmarks. On a high-complexity architecture search space for image classification, the Evo-NAS agent surpasses the accuracy achieved by commonly used agents with only 1/3 of the search cost.
Analyzing Pap cytology slides is an important tasks in detecting and grading precancerous and cancerous cervical cancer stages. Processing cytology images usually involve segmenting nuclei and overlapping cells. We introduce a cervical cytology dataset that can be used to evaluate nucleus detection, as well as image classification methods in the cytology image processing area. This dataset contains 93 real image stacks with their grade labels and manually annotated nuclei within images. We also present two methods: a baseline method based on a previously proposed approach, and a deep learning method, and compare their results with other state-of-the-art methods. Both the baseline method and the deep learning method outperform other state-of-the-art methods by significant margins. Along with the dataset, we publicly make the evaluation code and the baseline method available to download for further benchmarking.
In recent decade, many state-of-the-art algorithms on image classification as well as audio classification have achieved noticeable successes with the development of deep convolutional neural network (CNN). However, most of the works only exploit single type of training data. In this paper, we present a study on classifying bird species by exploiting the combination of both visual (images) and audio (sounds) data using CNN, which has been sparsely treated so far. Specifically, we propose CNN-based multimodal learning models in three types of fusion strategies (early, middle, late) to settle the issues of combining training data cross domains. The advantage of our proposed method lies on the fact that We can utilize CNN not only to extract features from image and audio data (spectrogram) but also to combine the features across modalities. In the experiment, we train and evaluate the network structure on a comprehensive CUB-200-2011 standard data set combing our originally collected audio data set with respect to the data species. We observe that a model which utilizes the combination of both data outperforms models trained with only an either type of data. We also show that transfer learning can significantly increase the classification performance.
New remote sensing sensors now acquire high spatial and spectral Satellite Image Time Series (SITS) of the world. These series of images are a key component of classification systems that aim at obtaining up-to-date and accurate land cover maps of the Earth's surfaces. More specifically, the combination of the temporal, spectral and spatial resolutions of new SITS makes possible to monitor vegetation dynamics. Although traditional classification algorithms, such as Random Forest (RF), have been successfully applied for SITS classification, these algorithms do not make the most of the temporal domain. Conversely, some approaches that take into account the temporal dimension have recently been tested, especially Recurrent Neural Networks (RNNs). This paper proposes an exhaustive study of another deep learning approaches, namely Temporal Convolutional Neural Networks (TempCNNs) where convolutions are applied in the temporal dimension. The goal is to quantitatively and qualitatively evaluate the contribution of TempCNNs for SITS classification. This paper proposes a set of experiments performed on one million time series extracted from 46 Formosat-2 images. The experimental results show that TempCNNs are more accurate than RF and RNNs, that are the current state of the art for SITS classification. We also highlight some differences with results obtained in computer vision, e.g. about pooling layers. Moreover, we provide some general guidelines on the network architecture, common regularization mechanisms, and hyper-parameter values such as batch size. Finally, we assess the visual quality of the land cover maps produced by TempCNNs.