1.Preparing for the Worst: Making Networks Less Brittle with Adversarial Batch Normalization ⬇️
Adversarial training is the industry standard for producing models that are robust to small adversarial perturbations. However, machine learning practitioners need models that are robust to domain shifts that occur naturally, such as changes in the style or illumination of input images. Such changes in input distribution have been effectively modeled as shifts in the mean and variance of deep image features. We adapt adversarial training by adversarially perturbing these feature statistics, rather than image pixels, to produce models that are robust to domain shift. We also visualize images from adversarially crafted distributions. Our method, Adversarial Batch Normalization (AdvBN), significantly improves the performance of ResNet-50 on ImageNet-C (+8.1%), Stylized-ImageNet (+6.7%), and ImageNet-Instagram (+3.9%) over standard training practices. In addition, we demonstrate that AdvBN can also improve generalization on semantic segmentation.
2.Light Direction and Color Estimation from Single Image with Deep Regression ⬇️
We present a method to estimate the direction and color of the scene light source from a single image. Our method is based on two main ideas: (a) we use a new synthetic dataset with strong shadow effects with similar constraints to the SID dataset; (b) we define a deep architecture trained on the mentioned dataset to estimate the direction and color of the scene light source. Apart from showing good performance on synthetic images, we additionally propose a preliminary procedure to obtain light positions of the Multi-Illumination dataset, and, in this way, we also prove that our trained model achieves good performance when it is applied to real scenes.
3.Multi-Resolution Graph Neural Network for Large-Scale Pointcloud Segmentation ⬇️
In this paper, we propose a multi-resolution deep-learning architecture to semantically segment dense large-scale pointclouds. Dense pointcloud data require a computationally expensive feature encoding process before semantic segmentation. Previous work has used different approaches to drastically downsample from the original pointcloud so common computing hardware can be utilized. While these approaches can relieve the computation burden to some extent, they are still limited in their processing capability for multiple scans. We present MuGNet, a memory-efficient, end-to-end graph neural network framework to perform semantic segmentation on large-scale pointclouds. We reduce the computation demand by utilizing a graph neural network on the preformed pointcloud graphs and retain the precision of the segmentation with a bidirectional network that fuses feature embedding at different resolutions. Our framework has been validated on benchmark datasets including Stanford Large-Scale 3D Indoor Spaces Dataset(S3DIS) and Virtual KITTI Dataset. We demonstrate that our framework can process up to 45 room scans at once on a single 11 GB GPU while still surpassing other graph-based solutions for segmentation on S3DIS with an 88.5% (+3%) overall accuracy and 69.8% (+7.7%) mIOU accuracy.
4.Deep Learning for 3D Point Cloud Understanding: A Survey ⬇️
The development of practical applications, such as autonomous driving and robotics, has brought increasing attention to 3D point cloud understanding. While deep learning has achieved remarkable success on image-based tasks, there are many unique challenges faced by deep neural networks in processing massive, unstructured and noisy 3D points. To demonstrate the latest progress of deep learning for 3D point cloud understanding, this paper summarizes recent remarkable research contributions in this area from several different directions (classification, segmentation, detection, tracking, flow estimation, registration, augmentation and completion), together with commonly used datasets, metrics and state-of-the-art performances. More information regarding this survey can be found at: this https URL.
5.Learning Unseen Emotions from Gestures via Semantically-Conditioned Zero-Shot Perception with Adversarial Autoencoders ⬇️
We present a novel generalized zero-shot algorithm to recognize perceived emotions from gestures. Our task is to map gestures to novel emotion categories not encountered in training. We introduce an adversarial, autoencoder-based representation learning that correlates 3D motion-captured gesture sequence with the vectorized representation of the natural-language perceived emotion terms using word2vec embeddings. The language-semantic embedding provides a representation of the emotion label space, and we leverage this underlying distribution to map the gesture-sequences to the appropriate categorical emotion labels. We train our method using a combination of gestures annotated with known emotion terms and gestures not annotated with any emotions. We evaluate our method on the MPI Emotional Body Expressions Database (EBEDB) and obtain an accuracy of
$58.43%$ . This improves the performance of current state-of-the-art algorithms for generalized zero-shot learning by$25$ --$27%$ on the absolute.
6.Image Captioning with Attention for Smart Local Tourism using EfficientNet ⬇️
Smart systems have been massively developed to help humans in various tasks. Deep Learning technologies push even further in creating accurate assistant systems due to the explosion of data lakes. One of the smart system tasks is to disseminate users needed information. This is crucial in the tourism sector to promote local tourism destinations. In this research, we design a model of local tourism specific image captioning, which later will support the development of AI-powered systems that assist various users. The model is developed using a visual Attention mechanism and uses the state-of-the-art feature extractor architecture EfficientNet. A local tourism dataset is collected and is used in the research, along with two different kinds of captions. Captions that describe the image literally and captions that represent human logical responses when seeing the image. This is done to make the captioning model more humane when implemented in the assistance system. We compared the performance of two different models using EfficientNet architectures (B0 and B4) with other well known VGG16 and InceptionV3. The best BLEU scores we get are 73.39 and 24.51 for the training set and the validation set respectively, using EfficientNetB0. The captioning result using the developed model shows that the model can produce logical caption for local tourism-related images
7.Faster Gradient-based NAS Pipeline Combining Broad Scalable Architecture with Confident Learning Rate ⬇️
In order to further improve the search efficiency of Neural Architecture Search (NAS), we propose B-DARTS, a novel pipeline combining broad scalable architecture with Confident Learning Rate (CLR). In B-DARTS, Broad Convolutional Neural Network (BCNN) is employed as the scalable architecture for DARTS, a popular differentiable NAS approach. On one hand, BCNN is a broad scalable architecture whose topology achieves two advantages compared with the deep one, mainly including faster single-step training speed and higher memory efficiency (i.e. larger batch size for architecture search), which are all contributed to the search efficiency improvement of NAS. On the other hand, DARTS discovers the optimal architecture by gradient-based optimization algorithm, which benefits from two superiorities of BCNN simultaneously. Similar to vanilla DARTS, B-DARTS also suffers from the performance collapse issue, where those weight-free operations are prone to be selected by the search strategy. Therefore, we propose CLR, that considers the confidence of gradient for architecture weights update increasing with the training time of over-parameterized model, to mitigate the above issue. Experimental results on CIFAR-10 and ImageNet show that 1) B-DARTS delivers state-of-the-art efficiency of 0.09 GPU day using first order approximation on CIFAR-10; 2) the learned architecture by B-DARTS achieves competitive performance using state-of-the-art composite multiply-accumulate operations and parameters on ImageNet; and 3) the proposed CLR is effective for performance collapse issue alleviation of both B-DARTS and DARTS.
8.PMVOS: Pixel-Level Matching-Based Video Object Segmentation ⬇️
Semi-supervised video object segmentation (VOS) aims to segment arbitrary target objects in video when the ground truth segmentation mask of the initial frame is provided. Due to this limitation of using prior knowledge about the target object, feature matching, which compares template features representing the target object with input features, is an essential step. Recently, pixel-level matching (PM), which matches every pixel in template features and input features, has been widely used for feature matching because of its high performance. However, despite its effectiveness, the information used to build the template features is limited to the initial and previous frames. We address this issue by proposing a novel method-PM-based video object segmentation (PMVOS)-that constructs strong template features containing the information of all past frames. Furthermore, we apply self-attention to the similarity maps generated from PM to capture global dependencies. On the DAVIS 2016 validation set, we achieve new state-of-the-art performance among real-time methods (> 30 fps), with a J&F score of 85.6%. Performance on the DAVIS 2017 and YouTube-VOS validation sets is also impressive, with J&F scores of 74.0% and 68.2%, respectively.
9.Synthetic Convolutional Features for Improved Semantic Segmentation ⬇️
Recently, learning-based image synthesis has enabled to generate high-resolution images, either applying popular adversarial training or a powerful perceptual loss. However, it remains challenging to successfully leverage synthetic data for improving semantic segmentation with additional synthetic images. Therefore, we suggest to generate intermediate convolutional features and propose the first synthesis approach that is catered to such intermediate convolutional features. This allows us to generate new features from label masks and include them successfully into the training procedure in order to improve the performance of semantic segmentation. Experimental results and analysis on two challenging datasets Cityscapes and ADE20K show that our generated feature improves performance on segmentation tasks.
10.IDA: Improved Data Augmentation Applied to Salient Object Detection ⬇️
In this paper, we present an Improved Data Augmentation (IDA) technique focused on Salient Object Detection (SOD). Standard data augmentation techniques proposed in the literature, such as image cropping, rotation, flipping, and resizing, only generate variations of the existing examples, providing a limited generalization. Our method combines image inpainting, affine transformations, and the linear combination of different generated background images with salient objects extracted from labeled data. Our proposed technique enables more precise control of the object's position and size while preserving background information. The background choice is based on an inter-image optimization, while object size follows a uniform random distribution within a specified interval, and the object position is intra-image optimal. We show that our method improves the segmentation quality when used for training state-of-the-art neural networks on several famous datasets of the SOD field. Combining our method with others surpasses traditional techniques such as horizontal-flip in 0.52% for F-measure and 1.19% for Precision. We also provide an evaluation in 7 different SOD datasets, with 9 distinct evaluation metrics and an average ranking of the evaluated methods.
11.Densely Guided Knowledge Distillation using Multiple Teacher Assistants ⬇️
With the success of deep neural networks, knowledge distillation which guides the learning of a small student network from a large teacher network is being actively studied for model compression and transfer learning. However, few studies have been performed to resolve the poor learning issue of the student network when the student and teacher model sizes significantly differ. In this paper, we propose a densely guided knowledge distillation using multiple teacher assistants that gradually decrease the model size to efficiently bridge the gap between teacher and student networks. To stimulate more efficient learning of the student network, we guide each teacher assistant to every other smaller teacher assistant step by step. Specifically, when teaching a smaller teacher assistant at the next step, the existing larger teacher assistants from the previous step are used as well as the teacher network to increase the learning efficiency. Moreover, we design stochastic teaching where, for each mini-batch during training, a teacher or a teacher assistant is randomly dropped. This acts as a regularizer like dropout to improve the accuracy of the student network. Thus, the student can always learn rich distilled knowledge from multiple sources ranging from the teacher to multiple teacher assistants. We verified the effectiveness of the proposed method for a classification task using Cifar-10, Cifar-100, and Tiny ImageNet. We also achieved significant performance improvements with various backbone architectures such as a simple stacked convolutional neural network, ResNet, and WideResNet.
12.$σ^2$R Loss: a Weighted Loss by Multiplicative Factors using Sigmoidal Functions ⬇️
In neural networks, the loss function represents the core of the learning process that leads the optimizer to an approximation of the optimal convergence error. Convolutional neural networks (CNN) use the loss function as a supervisory signal to train a deep model and contribute significantly to achieving the state of the art in some fields of artificial vision. Cross-entropy and Center loss functions are commonly used to increase the discriminating power of learned functions and increase the generalization performance of the model. Center loss minimizes the class intra-class variance and at the same time penalizes the long distance between the deep features inside each class. However, the total error of the center loss will be heavily influenced by the majority of the instances and can lead to a freezing state in terms of intra-class variance. To address this, we introduce a new loss function called sigma squared reduction loss ($\sigma^2$R loss), which is regulated by a sigmoid function to inflate/deflate the error per instance and then continue to reduce the intra-class variance. Our loss has clear intuition and geometric interpretation, furthermore, we demonstrate by experiments the effectiveness of our proposal on several benchmark datasets showing the intra-class variance reduction and overcoming the results obtained with center loss and soft nearest neighbour functions.
13.Commands 4 Autonomous Vehicles (C4AV) Workshop Summary ⬇️
The task of visual grounding requires locating the most relevant region or object in an image, given a natural language query. So far, progress on this task was mostly measured on curated datasets, which are not always representative of human spoken language. In this work, we deviate from recent, popular task settings and consider the problem under an autonomous vehicle scenario. In particular, we consider a situation where passengers can give free-form natural language commands to a vehicle which can be associated with an object in the street scene. To stimulate research on this topic, we have organized the \emph{Commands for Autonomous Vehicles} (C4AV) challenge based on the recent \emph{Talk2Car} dataset (URL: this https URL). This paper presents the results of the challenge. First, we compare the used benchmark against existing datasets for visual grounding. Second, we identify the aspects that render top-performing models successful, and relate them to existing state-of-the-art models for visual grounding, in addition to detecting potential failure cases by evaluating on carefully selected subsets. Finally, we discuss several possibilities for future work.
14.DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta ⬇️
Learning to generate new images for a novel category based on only a few images, named as few-shot image generation, has attracted increasing research interest. Several state-of-the-art works have yielded impressive results, but the diversity is still limited. In this work, we propose a novel Delta Generative Adversarial Network (DeltaGAN), which consists of a reconstruction subnetwork and a generation subnetwork. The reconstruction subnetwork captures intra-category transformation, i.e., "delta", between same-category pairs. The generation subnetwork generates sample-specific "delta" for an input image, which is combined with this input image to generate a new image within the same category. Besides, an adversarial delta matching loss is designed to link the above two subnetworks together. Extensive experiments on five few-shot image datasets demonstrate the effectiveness of our proposed method.
15.Moving object detection for visual odometry in a dynamic environment based on occlusion accumulation ⬇️
Detection of moving objects is an essential capability in dealing with dynamic environments. Most moving object detection algorithms have been designed for color images without depth. For robotic navigation where real-time RGB-D data is often readily available, utilization of the depth information would be beneficial for obstacle recognition.
Here, we propose a simple moving object detection algorithm that uses RGB-D images. The proposed algorithm does not require estimating a background model. Instead, it uses an occlusion model which enables us to estimate the camera pose on a background confused with moving objects that dominate the scene. The proposed algorithm allows to separate the moving object detection and visual odometry (VO) so that an arbitrary robust VO method can be employed in a dynamic situation with a combination of moving object detection, whereas other VO algorithms for a dynamic environment are inseparable. In this paper, we use dense visual odometry (DVO) as a VO method with a bi-square regression weight. Experimental results show the segmentation accuracy and the performance improvement of DVO in the situations. We validate our algorithm in public datasets and our dataset which also publicly accessible.
16.Contextual Semantic Interpretability ⬇️
Convolutional neural networks (CNN) are known to learn an image representation that captures concepts relevant to the task, but do so in an implicit way that hampers model interpretability. However, one could argue that such a representation is hidden in the neurons and can be made explicit by teaching the model to recognize semantically interpretable attributes that are present in the scene. We call such an intermediate layer a \emph{semantic bottleneck}. Once the attributes are learned, they can be re-combined to reach the final decision and provide both an accurate prediction and an explicit reasoning behind the CNN decision. In this paper, we look into semantic bottlenecks that capture context: we want attributes to be in groups of a few meaningful elements and participate jointly to the final decision. We use a two-layer semantic bottleneck that gathers attributes into interpretable, sparse groups, allowing them contribute differently to the final output depending on the context. We test our contextual semantic interpretable bottleneck (CSIB) on the task of landscape scenicness estimation and train the semantic interpretable bottleneck using an auxiliary database (SUN Attributes). Our model yields in predictions as accurate as a non-interpretable baseline when applied to a real-world test set of Flickr images, all while providing clear and interpretable explanations for each prediction.
17.Progressive Semantic-Aware Style Transformation for Blind Face Restoration ⬇️
Face restoration is important in face image processing, and has been widely studied in recent years. However, previous works often fail to generate plausible high quality (HQ) results for real-world low quality (LQ) face images. In this paper, we propose a new progressive semantic-aware style transformation framework, named PSFR-GAN, for face restoration. Specifically, instead of using an encoder-decoder framework as previous methods, we formulate the restoration of LQ face images as a multi-scale progressive restoration procedure through semantic-aware style transformation. Given a pair of LQ face image and its corresponding parsing map, we first generate a multi-scale pyramid of the inputs, and then progressively modulate different scale features from coarse-to-fine in a semantic-aware style transfer way. Compared with previous networks, the proposed PSFR-GAN makes full use of the semantic (parsing maps) and pixel (LQ images) space information from different scales of input pairs. In addition, we further introduce a semantic aware style loss which calculates the feature style loss for each semantic region individually to improve the details of face textures. Finally, we pretrain a face parsing network which can generate decent parsing maps from real-world LQ face images. Experiment results show that our model trained with synthetic data can not only produce more realistic high-resolution results for synthetic LQ inputs and but also generalize better to natural LQ face images compared with state-of-the-art methods. Codes are available at this https URL.
18.Learning Emotional-Blinded Face Representations ⬇️
We propose two face representations that are blind to facial expressions associated to emotional responses. This work is in part motivated by new international regulations for personal data protection, which enforce data controllers to protect any kind of sensitive information involved in automatic processes. The advances in Affective Computing have contributed to improve human-machine interfaces but, at the same time, the capacity to monitorize emotional responses triggers potential risks for humans, both in terms of fairness and privacy. We propose two different methods to learn these expression-blinded facial features. We show that it is possible to eliminate information related to emotion recognition tasks, while the performance of subject verification, gender recognition, and ethnicity classification are just slightly affected. We also present an application to train fairer classifiers in a case study of attractiveness classification with respect to a protected facial expression attribute. The results demonstrate that it is possible to reduce emotional information in the face representation while retaining competitive performance in other face-based artificial intelligence tasks.
19.Searching for Low-Bit Weights in Quantized Neural Networks ⬇️
Quantized neural networks with low-bit weights and activations are attractive for developing AI accelerators. However, the quantization functions used in most conventional quantization methods are non-differentiable, which increases the optimization difficulty of quantized networks. Compared with full-precision parameters (i.e., 32-bit floating numbers), low-bit values are selected from a much smaller set. For example, there are only 16 possibilities in 4-bit space. Thus, we present to regard the discrete weights in an arbitrary quantized neural network as searchable variables, and utilize a differential method to search them accurately. In particular, each weight is represented as a probability distribution over the discrete value set. The probabilities are optimized during training and the values with the highest probability are selected to establish the desired quantized network. Experimental results on benchmarks demonstrate that the proposed method is able to produce quantized neural networks with higher performance over the state-of-the-art methods on both image classification and super-resolution tasks.
20.DeepRemaster: Temporal Source-Reference Attention Networks for Comprehensive Video Enhancement ⬇️
The remastering of vintage film comprises of a diversity of sub-tasks including super-resolution, noise removal, and contrast enhancement which aim to restore the deteriorated film medium to its original state. Additionally, due to the technical limitations of the time, most vintage film is either recorded in black and white, or has low quality colors, for which colorization becomes necessary. In this work, we propose a single framework to tackle the entire remastering task semi-interactively. Our work is based on temporal convolutional neural networks with attention mechanisms trained on videos with data-driven deterioration simulation. Our proposed source-reference attention allows the model to handle an arbitrary number of reference color images to colorize long videos without the need for segmentation while maintaining temporal consistency. Quantitative analysis shows that our framework outperforms existing approaches, and that, in contrast to existing approaches, the performance of our framework increases with longer videos and more reference color images.
21.Conditional Image Generation with One-Vs-All Classifier ⬇️
This paper explores conditional image generation with a One-Vs-All classifier based on the Generative Adversarial Networks (GANs). Instead of the real/fake discriminator used in vanilla GANs, we propose to extend the discriminator to a One-Vs-All classifier (GAN-OVA) that can distinguish each input data to its category label. Specifically, we feed certain additional information as conditions to the generator and take the discriminator as a One-Vs-All classifier to identify each conditional category. Our model can be applied to different divergence or distances used to define the objective function, such as Jensen-Shannon divergence and Earth-Mover (or called Wasserstein-1) distance. We evaluate GAN-OVAs on MNIST and CelebA-HQ datasets, and the experimental results show that GAN-OVAs make progress toward stable training over regular conditional GANs. Furthermore, GAN-OVAs effectively accelerate the generation process of different classes and improves generation quality.
22.Face Sketch Synthesis with Style Transfer using Pyramid Column Feature ⬇️
In this paper, we propose a novel framework based on deep neural networks for face sketch synthesis from a photo. Imitating the process of how artists draw sketches, our framework synthesizes face sketches in a cascaded manner. A content image is first generated that outlines the shape of the face and the key facial features. Textures and shadings are then added to enrich the details of the sketch. We utilize a fully convolutional neural network (FCNN) to create the content image, and propose a style transfer approach to introduce textures and shadings based on a newly proposed pyramid column feature. We demonstrate that our style transfer approach based on the pyramid column feature can not only preserve more sketch details than the common style transfer method, but also surpasses traditional patch based methods. Quantitative and qualitative evaluations suggest that our framework outperforms other state-of-the-arts methods, and can also generalize well to different test images. Codes are available at this https URL
23.TopNet: Topology Preserving Metric Learning for Vessel Tree Reconstruction and Labelling ⬇️
Reconstructing Portal Vein and Hepatic Vein trees from contrast enhanced abdominal CT scans is a prerequisite for preoperative liver surgery simulation. Existing deep learning based methods treat vascular tree reconstruction as a semantic segmentation problem. However, vessels such as hepatic and portal vein look very similar locally and need to be traced to their source for robust label assignment. Therefore, semantic segmentation by looking at local 3D patch results in noisy misclassifications. To tackle this, we propose a novel multi-task deep learning architecture for vessel tree reconstruction. The network architecture simultaneously solves the task of detecting voxels on vascular centerlines (i.e. nodes) and estimates connectivity between center-voxels (edges) in the tree structure to be reconstructed. Further, we propose a novel connectivity metric which considers both inter-class distance and intra-class topological distance between center-voxel pairs. Vascular trees are reconstructed starting from the vessel source using the learned connectivity metric using the shortest path tree algorithm. A thorough evaluation on public IRCAD dataset shows that the proposed method considerably outperforms existing semantic segmentation based methods. To the best of our knowledge, this is the first deep learning based approach which learns multi-label tree structure connectivity from images.
24.Performance Monitoring of Object Detection During Deployment ⬇️
Performance monitoring of object detection is crucial for safety-critical applications such as autonomous vehicles that operate under varying and complex environmental conditions. Currently, object detectors are evaluated using summary metrics based on a single dataset that is assumed to be representative of all future deployment conditions. In practice, this assumption does not hold, and the performance fluctuates as a function of the deployment conditions. To address this issue, we propose an introspection approach to performance monitoring during deployment without the need for ground truth data. We do so by predicting when the per-frame mean average precision drops below a critical threshold using the detector's internal features. We quantitatively evaluate and demonstrate our method's ability to reduce risk by trading off making an incorrect decision by raising the alarm and absenting from detection.
25.Identification of Abnormal States in Videos of Ants Undergoing Social Phase Change ⬇️
Biology is both an important application area and a source of motivation for development of advanced machine learning techniques. Although much attention has been paid to large and complex data sets resulting from high-throughput sequencing, advances in high-quality video recording technology have begun to generate similarly rich data sets requiring sophisticated techniques from both computer vision and time-series analysis. Moreover, just as studying gene expression patterns in one organism can reveal general principles that apply to other organisms, the study of complex social interactions in an experimentally tractable model system, such as a laboratory ant colony, can provide general principles about the dynamics many other social groups. Here, we focus on one such example from the study of reproductive regulation in small laboratory colonies of $\sim$50 Harpgenathos ants. These ants can be artificially induced to begin a $\sim$20 day process of hierarchy reformation. Although the conclusion of this process is conspicuous to a human observer, it is still unclear which behaviors during the transients are contributing to the process. To address this issue, we explore the potential application of One-class Classification (OC) to the detection of abnormal states in ant colonies for which behavioral data is only available for the normal societal conditions during training. Specifically, we build upon the Deep Support Vector Data Description (DSVDD) and introduce the Inner-Outlier Generator (IO-GEN) that synthesizes fake "inner outlier" observations during training that are near the center of the DSVDD data description. We show that IO-GEN increases the reliability of the final OC classifier relative to other DSVDD baselines. This method can be used to screen video frames for which additional human observation is needed.
26.6-DoF Grasp Planning using Fast 3D Reconstruction and Grasp Quality CNN ⬇️
Recent consumer demand for home robots has accelerated performance of robotic grasping. However, a key component of the perception pipeline, the depth camera, is still expensive and inaccessible to most consumers. In addition, grasp planning has significantly improved recently, by leveraging large datasets and cloud robotics, and by limiting the state and action space to top-down grasps with 4 degrees of freedom (DoF). By leveraging multi-view geometry of the object using inexpensive equipment such as off-the-shelf RGB cameras and state-of-the-art algorithms such as Learn Stereo Machine (LSM\cite{kar2017learning}), the robot is able to generate more robust grasps from different angles with 6-DoF. In this paper, we present a modification of LSM to graspable objects, evaluate the grasps, and develop a 6-DoF grasp planner based on Grasp-Quality CNN (GQ-CNN\cite{mahler2017dex}) that exploits multiple camera views to plan a robust grasp, even in the absence of a possible top-down grasp.
27.Reinforcement Learning for Weakly Supervised Temporal Grounding of Natural Language in Untrimmed Videos ⬇️
Temporal grounding of natural language in untrimmed videos is a fundamental yet challenging multimedia task facilitating cross-media visual content retrieval. We focus on the weakly supervised setting of this task that merely accesses to coarse video-level language description annotation without temporal boundary, which is more consistent with reality as such weak labels are more readily available in practice. In this paper, we propose a \emph{Boundary Adaptive Refinement} (BAR) framework that resorts to reinforcement learning (RL) to guide the process of progressively refining the temporal boundary. To the best of our knowledge, we offer the first attempt to extend RL to temporal localization task with weak supervision. As it is non-trivial to obtain a straightforward reward function in the absence of pairwise granular boundary-query annotations, a cross-modal alignment evaluator is crafted to measure the alignment degree of segment-query pair to provide tailor-designed rewards. This refinement scheme completely abandons traditional sliding window based solution pattern and contributes to acquiring more efficient, boundary-flexible and content-aware grounding results. Extensive experiments on two public benchmarks Charades-STA and ActivityNet demonstrate that BAR outperforms the state-of-the-art weakly-supervised method and even beats some competitive fully-supervised ones.
28.Consistency Regularization with High-dimensional Non-adversarial Source-guided Perturbation for Unsupervised Domain Adaptation in Segmentation ⬇️
Unsupervised domain adaptation for semantic segmentation has been intensively studied due to the low cost of the pixel-level annotation for synthetic data. The most common approaches try to generate images or features mimicking the distribution in the target domain while preserving the semantic contents in the source domain so that a model can be trained with annotations from the latter. However, such methods highly rely on an image translator or feature extractor trained in an elaborated mechanism including adversarial training, which brings in extra complexity and instability in the adaptation process. Furthermore, these methods mainly focus on taking advantage of the labeled source dataset, leaving the unlabeled target dataset not fully utilized. In this paper, we propose a bidirectional style-induced domain adaptation method, called BiSIDA, that employs consistency regularization to efficiently exploit information from the unlabeled target domain dataset, requiring only a simple neural style transfer model. BiSIDA aligns domains by not only transferring source images into the style of target images but also transferring target images into the style of source images to perform high-dimensional perturbation on the unlabeled target images, which is crucial to the success in applying consistency regularization in segmentation tasks. Extensive experiments show that our BiSIDA achieves new state-of-the-art on two commonly-used synthetic-to-real domain adaptation benchmarks: GTA5-to-CityScapes and SYNTHIA-to-CityScapes.
29.Accelerating Search on Binary Codes in Weighted Hamming Space ⬇️
Compared to Hamming distance, weighted Hamming distance as a similarity measure between binary codes and the binary query point can provide superior accuracy in the search tasks. However, how to efficiently find
$K$ binary codes in the dataset that have the smallest weighted Hamming distance with the query is still an open issue. In this paper, a non-exhaustive search framework is proposed to accelerate the search speed and guarantee the search accuracy on the binary codes in weighted Hamming space. By separating the binary codes into multiple disjoint substrings as the bucket indices, the search framework iteratively probes the buckets until the query's nearest neighbors are found. The framework consists of two modules, the search module and the decision module. The search module successively probes the buckets and takes the candidates according to a proper probing sequence generated by the proposed search algorithm. And the decision module decides whether the query's nearest neighbors are found or more buckets should be probed according to a designed decision criterion. The analysis and experiments indicate that the search framework can solve the nearest neighbor search problem in weighted Hamming space and is orders of magnitude faster than the linear scan baseline.
30.MUTANT: A Training Paradigm for Out-of-Distribution Generalization in Visual Question Answering ⬇️
While progress has been made on the visual question answering leaderboards, models often utilize spurious correlations and priors in datasets under the i.i.d. setting. As such, evaluation on out-of-distribution (OOD) test samples has emerged as a proxy for generalization. In this paper, we present \textit{MUTANT}, a training paradigm that exposes the model to perceptually similar, yet semantically distinct \textit{mutations} of the input, to improve OOD generalization, such as the VQA-CP challenge. Under this paradigm, models utilize a consistency-constrained training objective to understand the effect of semantic changes in input (question-image pair) on the output (answer). Unlike existing methods on VQA-CP, \textit{MUTANT} does not rely on the knowledge about the nature of train and test answer distributions. \textit{MUTANT} establishes a new state-of-the-art accuracy on VQA-CP with a
$10.57%$ improvement. Our work opens up avenues for the use of semantic input mutations for OOD generalization in question answering.
31.Objective, Probabilistic, and Generalized Noise Level Dependent Classifications of sets of more or less 2D Periodic Images into Plane Symmetry Groups ⬇️
Crystallographic symmetry classifications from real-world images with periodicities in two dimensions (2D) are of interest to crystallographers and practitioners of computer vision studies alike. Currently, these classifications are typically made by both communities in a subjective manner that relies on arbitrary thresholds for judgments, and are reported under the pretense of being definitive, which is impossible. Moreover, the computer vision community tends to use direct space methods to make such classifications instead of more powerful and computationally efficient Fourier space methods. This is because the proper functioning of those methods requires more periodic repeats of a unit cell motif than are commonly present in images analyzed by the computer vision community. We demonstrate a novel approach to plane symmetry group classifications that is enabled by Kenichi Kanatani's Geometric Akaike Information Criterion and associated Geometric Akaike weights. Our approach leverages the advantages of working in Fourier space, is well suited for handling the hierarchic nature of crystallographic symmetries, and yields probabilistic results that are generalized noise level dependent. The latter feature means crystallographic symmetry classifications can be updated when less noisy image data and more accurate processing algorithms become available. We demonstrate the ability of our approach to objectively estimate the plane symmetry and pseudosymmetries of sets of synthetic 2D-periodic images with varying amounts of red-green-blue and spread noise. Additionally, we suggest a simple solution to the problem of too few periodic repeats in an input image for practical application of Fourier space methods. In doing so, we effectively solve the decades-old and heretofore intractable problem from computer vision of symmetry detection and classification from images in the presence of noise.
32.Smartphone Camera De-identification while Preserving Biometric Utility ⬇️
The principle of Photo Response Non Uniformity (PRNU) is often exploited to deduce the identity of the smartphone device whose camera or sensor was used to acquire a certain image. In this work, we design an algorithm that perturbs a face image acquired using a smartphone camera such that (a) sensor-specific details pertaining to the smartphone camera are suppressed (sensor anonymization); (b) the sensor pattern of a different device is incorporated (sensor spoofing); and (c) biometric matching using the perturbed image is not affected (biometric utility). We employ a simple approach utilizing Discrete Cosine Transform to achieve the aforementioned objectives. Experiments conducted on the MICHE-I and OULU-NPU datasets, which contain periocular and facial data acquired using 12 smartphone cameras, demonstrate the efficacy of the proposed de-identification algorithm on three different PRNU-based sensor identification schemes. This work has application in sensor forensics and personal privacy.
33.Predicting molecular phenotypes from histopathology images: a transcriptome-wide expression-morphology analysis in breast cancer ⬇️
Molecular phenotyping is central in cancer precision medicine, but remains costly and standard methods only provide a tumour average profile. Microscopic morphological patterns observable in histopathology sections from tumours are determined by the underlying molecular phenotype and associated with clinical factors. The relationship between morphology and molecular phenotype has a potential to be exploited for prediction of the molecular phenotype from the morphology visible in histopathology images.
We report the first transcriptome-wide Expression-MOrphology (EMO) analysis in breast cancer, where gene-specific models were optimised and validated for prediction of mRNA expression both as a tumour average and in spatially resolved manner. Individual deep convolutional neural networks (CNNs) were optimised to predict the expression of 17,695 genes from hematoxylin and eosin (HE) stained whole slide images (WSIs). Predictions for 9,334 (52.75%) genes were significantly associated with RNA-sequencing estimates (FDR adjusted p-value < 0.05). 1,011 of the genes were brought forward for validation, with 876 (87%) and 908 (90%) successfully replicated in internal and external test data, respectively. Predicted spatial intra-tumour variabilities in expression were validated in 76 genes, out of which 59 (77.6%) had a significant association (FDR adjusted p-value < 0.05) with spatial transcriptomics estimates. These results suggest that the proposed methodology can be applied to predict both tumour average gene expression and intra-tumour spatial expression directly from morphology, thus providing a scalable approach to characterise intra-tumour heterogeneity.
34.AdderSR: Towards Energy Efficient Image Super-Resolution ⬇️
This paper studies the single image super-resolution problem using adder neural networks (AdderNet). Compared with convolutional neural networks, AdderNet utilizing additions to calculate the output features thus avoid massive energy consumptions of conventional multiplications. However, it is very hard to directly inherit the existing success of AdderNet on large-scale image classification to the image super-resolution task due to the different calculation paradigm. Specifically, the adder operation cannot easily learn the identity mapping, which is essential for image processing tasks. In addition, the functionality of high-pass filters cannot be ensured by AdderNet. To this end, we thoroughly analyze the relationship between an adder operation and the identity mapping and insert shortcuts to enhance the performance of SR models using adder networks. Then, we develop a learnable power activation for adjusting the feature distribution and refining details. Experiments conducted on several benchmark models and datasets demonstrate that, our image super-resolution models using AdderNet can achieve comparable performance and visual quality to that of their CNN baselines with an about 2$\times$ reduction on the energy consumption.
35.Multi-modal Experts Network for Autonomous Driving ⬇️
End-to-end learning from sensory data has shown promising results in autonomous driving. While employing many sensors enhances world perception and should lead to more robust and reliable behavior of autonomous vehicles, it is challenging to train and deploy such network and at least two problems are encountered in the considered setting. The first one is the increase of computational complexity with the number of sensing devices. The other is the phenomena of network overfitting to the simplest and most informative input. We address both challenges with a novel, carefully tailored multi-modal experts network architecture and propose a multi-stage training procedure. The network contains a gating mechanism, which selects the most relevant input at each inference time step using a mixed discrete-continuous policy. We demonstrate the plausibility of the proposed approach on our 1/6 scale truck equipped with three cameras and one LiDAR.
36.Classification and Region Analysis of COVID-19 Infection using Lung CT Images and Deep Convolutional Neural Networks ⬇️
COVID-19 is a global health problem. Consequently, early detection and analysis of the infection patterns are crucial for controlling infection spread as well as devising a treatment plan. This work proposes a two-stage deep Convolutional Neural Networks (CNNs) based framework for delineation of COVID-19 infected regions in Lung CT images. In the first stage, initially, COVID-19 specific CT image features are enhanced using a two-level discrete wavelet transformation. These enhanced CT images are then classified using the proposed custom-made deep CoV-CTNet. In the second stage, the CT images classified as infectious images are provided to the segmentation models for the identification and analysis of COVID-19 infectious regions. In this regard, we propose a novel semantic segmentation model CoV-RASeg, which systematically uses average and max pooling operations in the encoder and decoder blocks. This systematic utilization of max and average pooling operations helps the proposed CoV-RASeg in simultaneously learning both the boundaries and region homogeneity. Moreover, the idea of attention is incorporated to deal with mildly infected regions. The proposed two-stage framework is evaluated on a standard Lung CT image dataset, and its performance is compared with the existing deep CNN models. The performance of the proposed CoV-CTNet is evaluated using Mathew Correlation Coefficient (MCC) measure (0.98) and that of proposed CoV-RASeg using Dice Similarity (DS) score (0.95). The promising results on an unseen test set suggest that the proposed framework has the potential to help the radiologists in the identification and analysis of COVID-19 infected regions.
37.Search and Rescue with Airborne Optical Sectioning ⬇️
We show that automated person detection under occlusion conditions can be significantly improved by combining multi-perspective images before classification. Here, we employed image integration by Airborne Optical Sectioning (AOS)---a synthetic aperture imaging technique that uses camera drones to capture unstructured thermal light fields---to achieve this with a precision/recall of 96/93%. Finding lost or injured people in dense forests is not generally feasible with thermal recordings, but becomes practical with use of AOS integral images. Our findings lay the foundation for effective future search and rescue technologies that can be applied in combination with autonomous or manned aircraft. They can also be beneficial for other fields that currently suffer from inaccurate classification of partially occluded people, animals, or objects.
38.Fused Deep Convolutional Neural Network for Precision Diagnosis of COVID-19 Using Chest X-Ray Images ⬇️
With a Coronavirus disease (COVID-19) case count exceeding 10 million worldwide, there is an increased need for a diagnostic capability. The main variables in increasing diagnostic capability are reduced cost, turnaround or diagnosis time, and upfront equipment cost and accessibility. Two candidates for machine learning COVID-19 diagnosis are Computed Tomography (CT) scans and plain chest X-rays. While CT scans score higher in sensitivity, they have a higher cost, maintenance requirement, and turnaround time as compared to plain chest X-rays. The use of portable chest X-radiograph (CXR) is recommended by the American College of Radiology (ACR) since using CT places a massive burden on radiology services. Therefore, X-ray imagery paired with machine learning techniques is proposed a first-line triage tool for COVID-19 diagnostics. In this paper we propose a computer-aided diagnosis (CAD) to accurately classify chest X-ray scans of COVID-19 and normal subjects by fine-tuning several neural networks (ResNet18, ResNet50, DenseNet201) pre-trained on the ImageNet dataset. These neural networks are fused in a parallel architecture and the voting criteria are applied in the final classification decision between the candidate object classes where the output of each neural network is representing a single vote. Several experiments are conducted on the weakly labeled COVID-19-CT-CXR dataset consisting of 263 COVID-19 CXR images extracted from PubMed Central Open Access subsets combined with 25 normal classification CXR images. These experiments show an optimistic result and a capability of the proposed model to outperforming many state-of-the-art algorithms on several measures. Using k-fold cross-validation and a bagging classifier ensemble, we achieve an accuracy of 99.7% and a sensitivity of 100%.
39.Residual Spatial Attention Network for Retinal Vessel Segmentation ⬇️
Reliable segmentation of retinal vessels can be employed as a way of monitoring and diagnosing certain diseases, such as diabetes and hypertension, as they affect the retinal vascular structure. In this work, we propose the Residual Spatial Attention Network (RSAN) for retinal vessel segmentation. RSAN employs a modified residual block structure that integrates DropBlock, which can not only be utilized to construct deep networks to extract more complex vascular features, but can also effectively alleviate the overfitting. Moreover, in order to further improve the representation capability of the network, based on this modified residual block, we introduce the spatial attention (SA) and propose the Residual Spatial Attention Block (RSAB) to build RSAN. We adopt the public DRIVE and CHASE DB1 color fundus image datasets to evaluate the proposed RSAN. Experiments show that the modified residual structure and the spatial attention are effective in this work, and our proposed RSAN achieves the state-of-the-art performance.
40.An Analysis by Synthesis Method that Allows Accurate Spatial Modeling of Thickness of Cortical Bone from Clinical QCT ⬇️
Osteoporosis is a skeletal disorder that leads to increased fracture risk due to decreased strength of cortical and trabecular bone. Even with state-of-the-art non-invasive assessment methods there is still a high underdiagnosis rate. Quantitative computed tomography (QCT) permits the selective analysis of cortical bone, however the low spatial resolution of clinical QCT leads to an overestimation of the thickness of cortical bone (this http URL) and bone strength.
We propose a novel, model based, fully automatic image analysis method that allows accurate spatial modeling of the thickness distribution of cortical bone from clinical QCT. In an analysis-by-synthesis (AbS) fashion a stochastic scan is synthesized from a probabilistic bone model, the optimal model parameters are estimated using a maximum a-posteriori approach. By exploiting the different characteristics of in-plane and out-of-plane point spread functions of CT scanners the proposed method is able assess the spatial distribution of cortical thickness.
The method was evaluated on eleven cadaveric human vertebrae, scanned by clinical QCT and analyzed using standard methods and AbS, both compared to high resolution peripheral QCT (HR-pQCT) as gold standard. While standard QCT based measurements overestimated this http URL. by 560% and did not show significant correlation with the gold standard ($r^2 = 0.20,, p = 0.169$ ) the proposed method eliminated the overestimation and showed a significant tight correlation with the gold standard ($r^2 = 0.98,, p < 0.0001$ ) a root mean square error below 10%.
41.Pruning Neural Networks at Initialization: Why are We Missing the Mark? ⬇️
Recent work has explored the possibility of pruning neural networks at initialization. We assess proposals for doing so: SNIP (Lee et al., 2019), GraSP (Wang et al., 2020), SynFlow (Tanaka et al., 2020), and magnitude pruning. Although these methods surpass the trivial baseline of random pruning, they remain below the accuracy of magnitude pruning after training, and we endeavor to understand why. We show that, unlike pruning after training, accuracy is the same or higher when randomly shuffling which weights these methods prune within each layer or sampling new initial values. As such, the per-weight pruning decisions made by these methods can be replaced by a per-layer choice of the fraction of weights to prune. This property undermines the claimed justifications for these methods and suggests broader challenges with the underlying pruning heuristics, the desire to prune at initialization, or both.
42.SCREENet: A Multi-view Deep Convolutional Neural Network for Classification of High-resolution Synthetic Mammographic Screening Scans ⬇️
Purpose: To develop and evaluate the accuracy of a multi-view deep learning approach to the analysis of high-resolution synthetic mammograms from digital breast tomosynthesis screening cases, and to assess the effect on accuracy of image resolution and training set size. Materials and Methods: In a retrospective study, 21,264 screening digital breast tomosynthesis (DBT) exams obtained at our institution were collected along with associated radiology reports. The 2D synthetic mammographic images from these exams, with varying resolutions and data set sizes, were used to train a multi-view deep convolutional neural network (MV-CNN) to classify screening images into BI-RADS classes (BI-RADS 0, 1 and 2) before evaluation on a held-out set of exams.
Results: Area under the receiver operating characteristic curve (AUC) for BI-RADS 0 vs non-BI-RADS 0 class was 0.912 for the MV-CNN trained on the full dataset. The model obtained accuracy of 84.8%, recall of 95.9% and precision of 95.0%. This AUC value decreased when the same model was trained with 50% and 25% of images (AUC = 0.877, P=0.010 and 0.834, P=0.009 respectively). Also, the performance dropped when the same model was trained using images that were under-sampled by 1/2 and 1/4 (AUC = 0.870, P=0.011 and 0.813, P=0.009 respectively).
Conclusion: This deep learning model classified high-resolution synthetic mammography scans into normal vs needing further workup using tens of thousands of high-resolution images. Smaller training data sets and lower resolution images both caused significant decrease in performance.
43.The Next Big Thing(s) in Unsupervised Machine Learning: Five Lessons from Infant Learning ⬇️
After a surge in popularity of supervised Deep Learning, the desire to reduce the dependence on curated, labelled data sets and to leverage the vast quantities of unlabelled data available recently triggered renewed interest in unsupervised learning algorithms. Despite a significantly improved performance due to approaches such as the identification of disentangled latent representations, contrastive learning, and clustering optimisations, the performance of unsupervised machine learning still falls short of its hypothesised potential. Machine learning has previously taken inspiration from neuroscience and cognitive science with great success. However, this has mostly been based on adult learners with access to labels and a vast amount of prior knowledge. In order to push unsupervised machine learning forward, we argue that developmental science of infant cognition might hold the key to unlocking the next generation of unsupervised learning approaches. Conceptually, human infant learning is the closest biological parallel to artificial unsupervised learning, as infants too must learn useful representations from unlabelled data. In contrast to machine learning, these new representations are learned rapidly and from relatively few examples. Moreover, infants learn robust representations that can be used flexibly and efficiently in a number of different tasks and contexts. We identify five crucial factors enabling infants' quality and speed of learning, assess the extent to which these have already been exploited in machine learning, and propose how further adoption of these factors can give rise to previously unseen performance levels in unsupervised learning.
44.Keep off the Grass: Permissible Driving Routes from Radar with Weak Audio Supervision ⬇️
Reliable outdoor deployment of mobile robots requires the robust identification of permissible driving routes in a given environment. The performance of LiDAR and vision-based perception systems deteriorates significantly if certain environmental factors are present e.g. rain, fog, darkness. Perception systems based on FMCW scanning radar maintain full performance regardless of environmental conditions and with a longer range than alternative sensors. Learning to segment a radar scan based on driveability in a fully supervised manner is not feasible as labelling each radar scan on a bin-by-bin basis is both difficult and time-consuming to do by hand. We therefore weakly supervise the training of the radar-based classifier through an audio-based classifier that is able to predict the terrain type underneath the robot. By combining odometry, GPS and the terrain labels from the audio classifier, we are able to construct a terrain labelled trajectory of the robot in the environment which is then used to label the radar scans. Using a curriculum learning procedure, we then train a radar segmentation network to generalise beyond the initial labelling and to detect all permissible driving routes in the environment.