1.Dissecting Person Re-identification from the Viewpoint of Viewpoint pdf
In person re-identification (re-ID),In person re-identification (re-ID), we usually refer the challenges of this task to variances in visual factors such as the viewpoint, pose, illumination and background. In spite of acknowledging these factors to be influential, quantitative studies on how they affect a re-ID system are still lacking.To gain insights in this scientific campaign, this paper makes an early attempt in studying a particular factor, viewpoint. We narrow the viewpoint problem down to the pedestrian rotation angle to obtain focused conclusions. In this regard, this paper makes two contributions to the community. First, we introduce a large-scale synthetic data engine, PersonX. Composed of hand-crafted 3D person models, the salient characteristic of this engine is "controllable". That is, we are able to synthesize pedestrians by setting the visual variables to arbitrary values. Second, on the 3D data engine, we quantitatively analyze the influence of pedestrian rotation angle on re-ID accuracy. Comprehensively, the person rotation angles are precisely customized from 0 to 360, allowing us to investigate its effect on the training, query, and gallery sets. Extensive experiment helps us gain deeper understanding of the fundamental problems in person re-ID. Our research also provides beneficial insights for dataset building and future practical usage, e.g., a person of a side view makes a better query.
2.Integrated unpaired appearance-preserving shape translation across domains pdf
We address the problem of un-supervised geometric image-to-image translation. Rather than transferring the style of an image as a whole, our goal is to translate the geometry of an object as depicted in different domains while preserving its appearance. Towards this goal, we propose a fully un-paired model that performs shape translation within a single model and without the need of additional post-processing stages. Extensive experiments on the VITON, CMU-Multi-PIE and our own FashionStyle datasets show the effectiveness of the proposed method at achieving the task at hand. In addition, we show that despite their low-dimensionality, the features learned by our model have potential for the item retrieval task
3.SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications pdf
One major factor impeding more widespread adoption of deep neural networks (DNNs) is their issues with robustness, which is essential for safety critical applications such as autonomous driving. This has motivated much recent work on adversarial attacks for DNNs, which mostly focus on pixel-level perturbations void of semantic meaning. In contrast, we present a general framework for adversarial black box attacks on agents, which are intimately related to the semantics of the task being performed by the agent. To do this, our proposed adversary (denoted as BBGAN) is trained to appropriately parametrize the environment (black box) with which the agent interacts, such that this agent performs poorly on its dedicated task. We illustrate the application of our BBGAN framework on three different tasks (primarily targeting aspects of autonomous navigation): object detection, self-driving, and autonomous UAV racing. On these tasks, our approach can be used to generate failure cases that fool an agent consistently.
4.Learning Attraction Field Representation for Robust Line Segment Detection pdf
This paper presents a region-partition based attraction field dual representation for line segment maps, and thus poses the problem of line segment detection (LSD) as the region coloring problem. The latter is then addressed by learning deep convolutional neural networks (ConvNets) for accuracy, robustness and efficiency. For a 2D line segment map, our dual representation consists of three components: (i) A region-partition map in which every pixel is assigned to one and only one line segment; (ii) An attraction field map in which every pixel in a partition region is encoded by its 2D projection vector w.r.t. the associated line segment; and (iii) A squeeze module which squashes the attraction field to a line segment map that almost perfectly recovers the input one. By leveraging the duality, we learn ConvNets to compute the attraction field maps for raw in-put images, followed by the squeeze module for LSD, in an end-to-end manner. Our method rigorously addresses several challenges in LSD such as local ambiguity and class imbalance. Our method also harnesses the best practices developed in ConvNets based semantic segmentation methods such as the encoder-decoder architecture and the a-trous convolution. In experiments, our method is tested on the WireFrame dataset and the YorkUrban dataset with state-of-the-art performance obtained. Especially, we advance the performance by 4.5 percents on the WireFrame dataset. Our method is also fast with 6.6~10.4 FPS, outperforming most of existing line segment detectors.
5.Understanding Individual Decisions of CNNs via Contrastive Backpropagation pdf
A number of backpropagation-based approaches such as DeConvNets, vanilla Gradient Visualization and Guided Backpropagation have been proposed to better understand individual decisions of deep convolutional neural networks. The saliency maps produced by them are proven to be non-discriminative. Recently, the Layer-wise Relevance Propagation (LRP) approach was proposed to explain the classification decisions of rectifier neural networks. In this work, we evaluate the discriminativeness of the generated explanations and analyze the theoretical foundation of LRP, i.e. Deep Taylor Decomposition. The experiments and analysis conclude that the explanations generated by LRP are not class-discriminative. Based on LRP, we propose Contrastive Layer-wise Relevance Propagation (CLRP), which is capable of producing instance-specific, class-discriminative, pixel-wise explanations. In the experiments, we use the CLRP to explain the decisions and understand the difference between neurons in individual classification decisions. We also evaluate the explanations quantitatively with a Pointing Game and an ablation study. Both qualitative and quantitative evaluations show that the CLRP generates better explanations than the LRP.
6.End-to-end Segmentation with Recurrent Attention Neural Network pdf
Image segmentation quality depends heavily on the quality of the image. For many medical imaging modalities, image reconstruction is required to convert acquired raw data to images before any analysis. However, imperfect reconstruction with artifacts and loss of information is almost inevitable, which compromises the final performance of segmentation. In this study, we present a novel end-to-end deep learning framework that performs magnetic resonance brain image segmentation directly from the raw data. The end-to-end framework consists a unique task-driven attention module that recurrently utilizes intermediate segmentation result to facilitate image-domain feature extraction from the raw data for segmentation, thus closely bridging the reconstruction and the segmentation tasks. In addition, we introduce a novel workflow to generate labeled training data for segmentation by exploiting imaging modality simulators and digital phantoms. Extensive experiment results show that the proposed method outperforms the state-of-the-art methods.
7.Point-to-Pose Voting based Hand Pose Estimation using Residual Permutation Equivariant Layer pdf
Recently, 3D input data based hand pose estimation methods have shown state-of-the-art performance, because 3D data capture more spatial information than the depth image. Whereas 3D voxel-based methods need a large amount of memory, PointNet based methods need tedious preprocessing steps such as K-nearest neighbour search for each point. In this paper, we present a novel deep learning hand pose estimation method for an unordered point cloud. Our method takes 1024 3D points as input and does not require additional information. We use Permutation Equivariant Layer (PEL) as the basic element, where a residual network version of PEL is proposed for the hand pose estimation task. Furthermore, we propose a voting based scheme to merge information from individual points to the final pose output. In addition to the pose estimation task, the voting-based scheme can also provide point cloud segmentation result without ground-truth for segmentation. We evaluate our method on both NYU dataset and the Hands2017Challenge dataset. Our method outperforms recent state-of-the-art methods, where our pose accuracy is currently the best for the Hands2017Challenge dataset.
8.Learn to See by Events: RGB Frame Synthesis from Event Cameras pdf
Event cameras are biologically-inspired sensors that gather the temporal evolution of the scene, capturing only pixel-wise brightness variations. Despite having multiple advantages with respect to traditional cameras, their use is still limited due to the difficult intelligibility and restricted usability through traditional vision algorithms. To this aim, we present a framework which exploits the output of event cameras to synthesize RGB frames. In particular, the frame generation relies on an initial or a periodic set of color key-frames and a sequence of intermediate event frames, i.e. gray-level images that integrate the brightness changes captured by the event camera during a short temporal slot. An adversarial architecture combined with a recurrent module is employed for the frame synthesis. Both traditional and event-based datasets are adopted to assess the capabilities of the proposed architecture: pixel-wise and semantic metrics confirm the quality of the synthesized images.
9.Dynamic Spatio-temporal Graph-based CNNs for Traffic Prediction pdf
Accurate traffic forecast is a challenging problem due to the large-scale problem size, as well as the complex and dynamic nature of spatio-temporal dependency of traffic flow. Most existing graph-based CNNs attempt to capture the static relations while largely neglecting the dynamics underlying sequential data. In this paper, we present dynamic spatio-temporal graph-based CNNs (DST-GCNNs) by learning expressive features to represent spatio-temporal structures and predict future traffic from historical traffic flow. In particular, DST-GCNN is a two stream network. In the flow prediction stream, we present a novel graph-based spatio-temporal convolutional layer to extract features from a graph representation of traffic flow. Then several such layers are stacked together to predict future traffic over time. Meanwhile, the proximity relations between nodes in the graph are often time variant as the traffic condition changes over time. To capture the graph dynamics, we use the graph prediction stream to predict the dynamic graph structures, and the predicted structures are fed into the flow prediction stream. Experiments on real traffic datasets demonstrate that the proposed model achieves competitive performances compared with the other state-of-the-art methods.
10.VideoMem: Constructing, Analyzing, Predicting Short-term and Long-term Video Memorability pdf
Humans share a strong tendency to memorize/forget some of the visual information they encounter. This paper focuses on providing computational models for the prediction of the intrinsic memorability of visual content. To address this new challenge, we introduce a large scale dataset (VideoMem) composed of 10,000 videos annotated with memorability scores. In contrast to previous work on image memorability -- where memorability was measured a few minutes after memorization -- memory performance is measured twice: a few minutes after memorization and again 24-72 hours later. Hence, the dataset comes with short-term and long-term memorability annotations. After an in-depth analysis of the dataset, we investigate several deep neural network based models for the prediction of video memorability. Our best model using a ranking loss achieves a Spearman's rank correlation of 0.494 for short-term memorability prediction, while our proposed model with attention mechanism provides insights of what makes a content memorable. The VideoMem dataset with pre-extracted features is publicly available.
11.Summarizing Videos with Attention pdf
In this work we propose a novel method for supervised, keyshots based video summarization by applying a conceptually simple and computationally efficient soft, self-attention mechanism. Current state of the art methods leverage bi-directional recurrent networks such as BiLSTM combined with attention. These networks are complex to implement and computationally demanding compared to fully connected networks. To that end we propose a simple, self-attention based network for video summarization which performs the entire sequence to sequence transformation in a single feed forward pass and single backward pass during training. Our method sets a new state of the art results on two benchmarks TvSum and SumMe, commonly used in this domain.
12.Unsupervised Generation of Optical Flow Datasets from Videos in the Wild pdf
Dense optical flow ground truth of non-rigid motion for real-world images are not available due to the non-intuitive annotation. Aiming at training optical flow deep networks, we present an unsupervised algorithm to generate optical flow ground truth from real-world videos. The algorithm extracts and matches objects of interest from pairs of images in videos to find initial constraints, and applies as-rigid-as-possible deformation over the objects of interest to obtain dense flow fields. The ground truth correctness is enforced by warping the objects in the first frames using the flow fields. We apply the algorithm on the DAVIS dataset to obtain optical flow ground truths for non-rigid movement of real-world objects, using either ground truth or predicted segmentation. We discuss several methods to increase the optical flow variations in the dataset. Extensive experimental results show that training on non-rigid real motion is beneficial compared to training on rigid synthetic data. Moreover, we show that our pipeline generates training data suitable to train successfully FlowNet-S, PWC-Net, and LiteFlowNet deep networks.
13.Stacked Dense U-Nets with Dual Transformers for Robust Face Alignment pdf
Facial landmark localisation in images captured in-the-wild is an important and challenging problem. The current state-of-the-art revolves around certain kinds of Deep Convolutional Neural Networks (DCNNs) such as stacked U-Nets and Hourglass networks. In this work, we innovatively propose stacked dense U-Nets for this task. We design a novel scale aggregation network topology structure and a channel aggregation building block to improve the model's capacity without sacrificing the computational complexity and model size. With the assistance of deformable convolutions inside the stacked dense U-Nets and coherent loss for outside data transformation, our model obtains the ability to be spatially invariant to arbitrary input face images. Extensive experiments on many in-the-wild datasets, validate the robustness of the proposed method under extreme poses, exaggerated expressions and heavy occlusions. Finally, we show that accurate 3D face alignment can assist pose-invariant face recognition where we achieve a new state-of-the-art accuracy on CFP-FP.
14.An Empirical Study towards Understanding How Deep Convolutional Nets Recognize Falls pdf
Detecting unintended falls is essential for ambient intelligence and healthcare of elderly people living alone. In recent years, deep convolutional nets are widely used in human action analysis, based on which a number of fall detection methods have been proposed. Despite their highly effective performances, the behaviors of how the convolutional nets recognize falls are still not clear. In this paper, instead of proposing a novel approach, we perform a systematical empirical study, attempting to investigate the underlying fall recognition process. We propose four tasks to investigate, which involve five types of input modalities, seven net instances and different training samples. The obtained quantitative and qualitative results reveal the patterns that the nets tend to learn, and several factors that can heavily influence the performances on fall recognition. We expect that our conclusions are favorable to proposing better deep learning solutions to fall detection systems.
15.Local Temporal Bilinear Pooling for Fine-grained Action Parsing pdf
Fine-grained temporal action parsing is important in many applications, such as daily activity understanding, human motion analysis, surgical robotics and others requiring subtle and precise operations in a long-term period. In this paper we propose a novel bilinear pooling operation, which is used in intermediate layers of a temporal convolutional encoder-decoder net. In contrast to other work, our proposed bilinear pooling is learnable and hence can capture more complex local statistics than the conventional counterpart. In addition, we introduce exact lower-dimension representations of our bilinear forms, so that the dimensionality is reduced with neither information loss nor extra computation. We perform intensive experiments to quantitatively analyze our model and show the superior performances to other state-of-the-art work on various datasets.
16.Learning to generate filters for convolutional neural networks pdf
Conventionally, convolutional neural networks (CNNs) process different images with the same set of filters. However, the variations in images pose a challenge to this fashion. In this paper, we propose to generate sample-specific filters for convolutional layers in the forward pass. Since the filters are generated on-the-fly, the model becomes more flexible and can better fit the training data compared to traditional CNNs. In order to obtain sample-specific features, we extract the intermediate feature maps from an autoencoder. As filters are usually high dimensional, we propose to learn a set of coefficients instead of a set of filters. These coefficients are used to linearly combine the base filters from a filter repository to generate the final filters for a CNN. The proposed method is evaluated on MNIST, MTFL and CIFAR10 datasets. Experiment results demonstrate that the classification accuracy of the baseline model can be improved by using the proposed filter generation method.
17.Interactive Full Image Segmentation pdf
We address the task of interactive full image annotation, where the goal is to produce accurate segmentations for all object and stuff regions in an image. To this end we propose an interactive, scribble-based annotation framework which operates on the whole image to produce segmentations for all regions. This enables the annotator to focus on the largest errors made by the machine across the whole image, and to share corrections across nearby regions. Furthermore, we adapt Mask-RCNN into a fast interactive segmentation framework and introduce a new instance-aware loss measured at the pixel-level in the full image canvas, which lets predictions for nearby regions properly compete. Finally, we compare to interactive single object segmentation on the on the COCO panoptic dataset. We demonstrate that, at a budget of four extreme clicks and four corrective scribbles per region, our interactive full image segmentation approach leads to a 5% IoU gain, reaching 90% IoU.
18.Learning to Compose Dynamic Tree Structures for Visual Contexts pdf
We propose to compose dynamic tree structures that place the objects in an image into a visual context, helping visual reasoning tasks such as scene graph generation and visual Q&A. Our visual context tree model, dubbed VCTree, has two key advantages over existing structured object representations including chains and fully-connected graphs: 1) The efficient and expressive binary tree encodes the inherent parallel/hierarchical relationships among objects, e.g., "clothes" and "pants" are usually co-occur and belong to "person"; 2) the dynamic structure varies from image to image and task to task, allowing more content-/task-specific message passing among objects. To construct a VCTree, we design a score function that calculates the task-dependent validity between each object pair, and the tree is the binary version of the maximum spanning tree from the score matrix. Then, visual contexts are encoded by bidirectional TreeLSTM and decoded by task-specific models. We develop a hybrid learning procedure which integrates end-task supervised learning and the tree structure reinforcement learning, where the former's evaluation result serves as a self-critic for the latter's structure exploration. Experimental results on two benchmarks, which require reasoning over contexts: Visual Genome for scene graph generation and VQA2.0 for visual Q&A, show that VCTree outperforms state-of-the-art results while discovering interpretable visual context structures.
19.Few-shot Object Detection via Feature Reweighting pdf
This work aims to solve the challenging few-shot object detection problem where only a few annotated examples are available for each object category to train a detection model. Such an ability of learning to detect an object from just a few examples is common for human vision systems, but remains absent for computer vision systems. Though few-shot meta learning offers a promising solution technique, previous works mostly target the task of image classification and are not directly applicable for the much more complicated object detection task. In this work, we propose a novel meta-learning based model with carefully designed architecture, which consists of a meta-model and a base detection model. The base detection model is trained on several base classes with sufficient samples to offer basis features. The meta-model is trained to reweight importance of features from the base detection model over the input image and adapt these features to assist novel object detection from a few examples. The meta-model is light-weight, end-to-end trainable and able to entail the base model with detection ability for novel objects fast. Through experiments we demonstrated our model can outperform baselines by a large margin for few-shot object detection, on multiple datasets and settings. Our model also exhibits fast adaptation speed to novel few-shot classes.
20.Enhancing Label-Driven Deep Deformable Image Registration with Local Distance Metrics for State-of-the-Art Cardiac Motion Tracking pdf
While deep learning has achieved significant advances in accuracy for medical image segmentation, its benefits for deformable image registration have so far remained limited to reduced computation times. Previous work has either focused on replacing the iterative optimization of distance and smoothness terms with CNN-layers or using supervised approaches driven by labels. Our method is the first to combine the complementary strengths of global semantic information (represented by segmentation labels) and local distance metrics that help align surrounding structures. We demonstrate significant higher Dice scores (of 86.5%) for deformable cardiac image registration compared to classic registration (79.0%) as well as label-driven deep learning frameworks (83.4%).
21.Explainable and Explicit Visual Reasoning over Scene Graphs pdf
We aim to dismantle the prevalent black-box neural architectures used in complex visual reasoning tasks, into the proposed eXplainable and eXplicit Neural Modules (XNMs), which advance beyond existing neural module networks towards using scene graphs --- objects as nodes and the pairwise relationships as edges --- for explainable and explicit reasoning with structured knowledge. XNMs allow us to pay more attention to teach machines how to "think", regardless of what they "look". As we will show in the paper, by using scene graphs as an inductive bias, 1) we can design XNMs in a concise and flexible fashion, i.e., XNMs merely consist of 4 meta-types, which significantly reduce the number of parameters by 10 to 100 times, and 2) we can explicitly trace the reasoning-flow in terms of graph attentions. XNMs are so generic that they support a wide range of scene graph implementations with various qualities. For example, when the graphs are detected perfectly, XNMs achieve 100% accuracy on both CLEVR and CLEVR CoGenT, establishing an empirical performance upper-bound for visual reasoning; when the graphs are noisily detected from real-world images, XNMs are still robust to achieve a competitive 67.5% accuracy on VQAv2.0, surpassing the popular bag-of-objects attention models without graph structures.
22.Feature Matters: A Stage-by-Stage Approach for Knowledge Transfer pdf
Convolutional Neural Networks (CNNs) become deeper and deeper in recent years, making the study of model acceleration imperative. It is a common practice to employ a shallow network, called student, to learn from a deep one, which is termed as teacher. Prior work made many attempts to transfer different types of knowledge from teacher to student, however, there are two problems remaining unsolved. Firstly, the knowledge used by existing methods is usually manually defined, which may not be consistent with the information learned by the original model. Secondly, there lacks an effective training scheme for the transfer process, leading to degradation of performance. In this work, we argue that feature is the most important knowledge from teacher. It is sufficient for student to achieve appealing performance by just learning similar features as teacher without any processing. Based on this discovery, we further present an efficient learning strategy, which is to make student mimic features of teacher stage by stage. Extensive experiments suggest that the proposed approach significantly narrows down the gap between student and teacher, and shows strong stability on various tasks, ie classification and detection, outperforming the state-of-the-art methods.
23.Visual Attention for Behavioral Cloning in Autonomous Driving pdf
The goal of our work is to use visual attention to enhance autonomous driving performance. We present two methods of predicting visual attention maps. The first method is a supervised learning approach in which we collect eye-gaze data for the task of driving and use this to train a model for predicting the attention map. The second method is a novel unsupervised approach where we train a model to learn to predict attention as it learns to drive a car. Finally, we present a comparative study of our results and show that the supervised approach for predicting attention when incorporated performs better than other approaches.
24.Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders pdf
Many approaches in generalized zero-shot learning rely on cross-modal mapping between the image feature space and the class embedding space. As labeled images are rare, one direction is to augment the dataset by generating either images or image features. However, the former misses fine-grained details and the latter requires learning a mapping associated with class embeddings. In this work, we take feature generation one step further and propose a model where a shared latent space of image features and class embeddings is learned by modality-specific aligned variational autoencoders. This leaves us with the required discriminative information about the image and classes in the latent features, on which we train a softmax classifier. The key to our approach is that we align the distributions learned from images and from side-information to construct latent features that contain the essential multi-modal information associated with unseen classes. We evaluate our learned latent features on several benchmark datasets, i.e. CUB, SUN, AWA1 and AWA2, and establish a new state-of-the-art on generalized zero-shot as well as on few-shot learning. Moreover, our results on ImageNet with various zero-shot splits show that our latent features generalize well in large-scale settings.
25.Capture Dense: Markerless Motion Capture Meets Dense Pose Estimation pdf
We present a method to combine markerless motion capture and dense pose feature estimation into a single framework. We demonstrate that dense pose information can help for multiview/single-view motion capture, and multiview motion capture can help the collection of a high-quality dataset for training the dense pose detector. Specifically, we first introduce a novel markerless motion capture method that can take advantage of dense parsing capability provided by the dense pose detector. Thanks to the introduced dense human parsing ability, our method is demonstrated much more efficient, and accurate compared with the available state-of-the-art markerless motion capture approach. Second, we improve the performance of available dense pose detector by using multiview markerless motion capture data. Such dataset is beneficial to dense pose training because they are more dense and accurate and consistent, and can compensate for the corner cases such as unusual viewpoints. We quantitatively demonstrate the improved performance of our dense pose detector over the available DensePose. Our dense pose dataset and detector will be made public.
26.Multi$^{\mathbf{3}}$Net: Segmenting Flooded Buildings via Fusion of Multiresolution, Multisensor, and Multitemporal Satellite Imagery pdf
We propose a novel approach for rapid segmentation of flooded buildings by fusing multiresolution, multisensor, and multitemporal satellite imagery in a convolutional neural network. Our model significantly expedites the generation of satellite imagery-based flood maps, crucial for first responders and local authorities in the early stages of flood events. By incorporating multitemporal satellite imagery, our model allows for rapid and accurate post-disaster damage assessment and can be used by governments to better coordinate medium- and long-term financial assistance programs for affected areas. The network consists of multiple streams of encoder-decoder architectures that extract spatiotemporal information from medium-resolution images and spatial information from high-resolution images before fusing the resulting representations into a single medium-resolution segmentation map of flooded buildings. We compare our model to state-of-the-art methods for building footprint segmentation as well as to alternative fusion approaches for the segmentation of flooded buildings and find that our model performs best on both tasks. We also demonstrate that our model produces highly accurate segmentation maps of flooded buildings using only publicly available medium-resolution data instead of significantly more detailed but sparsely available very high-resolution data. We release the first open-source dataset of fully preprocessed and labeled multiresolution, multispectral, and multitemporal satellite images of disaster sites along with our source code.
27.Moment Matching for Multi-Source Domain Adaptation pdf
Conventional unsupervised domain adaptation (UDA) assumes that training data are sampled from a single domain. This neglects the more practical scenario where training data are collected from multiple sources, requiring multi-source domain adaptation. We make three major contributions towards addressing this problem. First, we propose a new deep learning approach, Moment Matching for Multi-Source Domain Adaptation M3SDA, which aims to transfer knowledge learned from multiple labeled source domains to an unlabeled target domain by dynamically aligning moments of their feature distributions. Second, we provide a sound theoretical analysis of moment-related error bounds for multi-source domain adaptation. Third, we collect and annotate by far the largest UDA dataset with six distinct domains and approximately 0.6 million images distributed among 345 categories, addressing the gap in data availability for multi-source UDA research. Extensive experiments are performed to demonstrate the effectiveness of our proposed model, which outperforms existing state-of-the-art methods by a large margin.
28.Cerebrovascular Network Segmentation on MRA Images with Deep Learning pdf
Deep learning has been shown to produce state of the art results in many tasks in biomedical imaging, especially in segmentation. Moreover, segmentation of the cerebrovascular structure from magnetic resonance angiography is a challenging problem because its complex geometry and topology have a large inter-patient variability. Therefore, in this work, we present a convolutional neural network approach for this problem. Particularly, a new network topology inspired by the U-net 3D and by the Inception modules, entitled Uception. In addition, a discussion about the best objective function for sparse data also guided most choices during the project. State of the art models are also implemented for a comparison purpose and final results show that the proposed architecture has the best performance in this particular context.
29.Complete the Look: Scene-based Complementary Product Recommendation pdf
Modeling fashion compatibility is challenging due to its complexity and subjectivity. Existing work focuses on predicting compatibility between product images (e.g. an image containing a t-shirt and an image containing a pair of jeans). However, these approaches ignore real-world 'scene' images (e.g. selfies); such images are hard to deal with due to their complexity, clutter, variations in lighting and pose (etc.) but on the other hand could potentially provide key context (e.g. the user's body type, or the season) for making more accurate recommendations. In this work, we propose a new task called 'Complete the Look', which seeks to recommend visually compatible products based on scene images. We design an approach to extract training data for this task, and propose a novel way to learn the scene-product compatibility from fashion or interior design images. Our approach measures compatibility both globally and locally via CNNs and attention mechanisms. Extensive experiments show that our method achieves significant performance gains over alternative systems. Human evaluation and qualitative analysis are also conducted to further understand model behavior. We hope this work could lead to useful applications which link large corpora of real-world scenes with shoppable products.
30.Learning Single-View 3D Reconstruction with Adversarial Training pdf
Single-view 3D shape reconstruction is an important but challenging problem, mainly for two reasons. First, as shape annotation is very expensive to acquire, current methods rely on synthetic data, in which ground-truth 3D annotation is easy to obtain. However, this results in domain adaptation problem when applied to natural images. The second challenge is that it exists multiple shapes that can explain a given 2D image. In this paper, we propose a framework to improve over these challenges using adversarial training. On one hand, we impose domain-confusion between natural and synthetic image representations to reduce the distribution gap. On the other hand, we impose the reconstruction to be `realistic' by forcing it to lie on a (learned) manifold of realistic object shapes. Moreover, our experiments show that these constraints improve performance by a large margin over a baseline reconstruction model. We achieve results competitive with the state of the art using only RGB images and with a much simpler architecture.
31.Multiview Cross-supervision for Semantic Segmentation pdf
This paper presents a semi-supervised learning framework for a customized semantic segmentation task using multiview image streams. A key challenge of the customized task lies in the limited accessibility of the labeled data due to the requirement of prohibitive manual annotation effort. We hypothesize that it is possible to leverage multiview image streams that are linked through the underlying 3D geometry, which can provide an additional supervisionary signal to train a segmentation model. We formulate a new cross-supervision method using a shape belief transfer---the segmentation belief in one image is used to predict that of the other image through epipolar geometry analogous to shape-from-silhouette. The shape belief transfer provides the upper and lower bounds of the segmentation for the unlabeled data where its gap approaches asymptotically to zero as the number of the labeled views increases. We integrate this theory to design a novel network that is agnostic to camera calibration, network model, and semantic category and bypasses the intermediate process of suboptimal 3D reconstruction. We validate this network by recognizing a customized semantic category per pixel from realworld visual data including non-human species and a subject of interest in social videos where attaining large-scale annotation data is infeasible.
32.Decompose to manipulate: Manipulable Object Synthesis in 3D Medical Images with Structured Image Decomposition pdf
The performance of medical image analysis systems is constrained by the quantity of high-quality image annotations. Such systems require data to be annotated by experts with years of training, especially when diagnostic decisions are involved. Such datasets are thus hard to scale up. In this context, it is hard for supervised learning systems to generalize to the cases that are rare in the training set but would be present in real-world clinical practices. We believe that the synthetic image samples generated by a system trained on the real data can be useful for improving the supervised learning tasks in the medical image analysis applications. Allowing the image synthesis to be manipulable could help synthetic images provide complementary information to the training data rather than simply duplicating the real-data manifold. In this paper, we propose a framework for synthesizing 3D objects, such as pulmonary nodules, in 3D medical images with manipulable properties. The manipulation is enabled by decomposing of the object of interests into its segmentation mask and a 1D vector containing the residual information. The synthetic object is refined and blended into the image context with two adversarial discriminators. We evaluate the proposed framework on lung nodules in 3D chest CT images and show that the proposed framework could generate realistic nodules with manipulable shapes, textures and locations, etc. By sampling from both the synthetic nodules and the real nodules from 2800 3D CT volumes during the classifier training, we show the synthetic patches could improve the overall nodule detection performance by average 8.44% competition performance metric (CPM) score.
33.Knowing what you know in brain segmentation using deep neural networks pdf
In this paper, we describe a deep neural network trained to predict FreeSurfer segmentations of structural MRI volumes, in seconds rather than hours. The network was trained and evaluated on the largest dataset ever assembled for this purpose, obtained by combining data from more than a hundred sites. We also show that the prediction uncertainty of the network at each voxel is a good indicator of whether the network has made an error. The resulting uncertainty volume can be used in conjunction with the predicted segmentation to improve downstream uses, such as calculation of measures derived from segmentation regions of interest or the building of prediction models. Finally, we demonstrate that the average prediction uncertainty across voxels in the brain is an excellent indicator of manual quality control ratings, outperforming the best available automated solutions.
34.Deep Learning for Classical Japanese Literature pdf
Much of machine learning research focuses on producing models which perform well on benchmark tasks, in turn improving our understanding of the challenges associated with those tasks. From the perspective of ML researchers, the content of the task itself is largely irrelevant, and thus there have increasingly been calls for benchmark tasks to more heavily focus on problems which are of social or cultural relevance. In this work, we introduce Kuzushiji-MNIST, a dataset which focuses on Kuzushiji (cursive Japanese), as well as two larger, more challenging datasets, Kuzushiji-49 and Kuzushiji-Kanji. Through these datasets, we wish to engage the machine learning community into the world of classical Japanese literature. Dataset available at this https URL
35.Towards Accurate Generative Models of Video: A New Metric & Challenges pdf
Recent advances in deep generative models have lead to remarkable progress in synthesizing high quality images. Following their successful application in image processing and representation learning, an important next step is to consider videos. Learning generative models of video is a much harder task, requiring a model to capture the temporal dynamics of a scene, in addition to the visual presentation of objects. Although recent attempts at formulating generative models of video have had some success, current progress is hampered by (1) the lack of qualitative metrics that consider visual quality, temporal coherence, and diversity of samples, and (2) the wide gap between purely synthetic video datasets and challenging real-world datasets in terms of complexity. To this extent we propose Fréchet Video Distance (FVD), a new metric for generative models of video based on FID, and StarCraft 2 Videos (SCV), a collection of progressively harder datasets that challenge the capabilities of the current iteration of generative models for video. We conduct a large-scale human study, which confirms that FVD correlates well with qualitative human judgment of generated videos, and provide initial benchmark results on SCV.
36.Learning to Unlearn: Building Immunity to Dataset Bias in Medical Imaging Studies pdf
Medical imaging machine learning algorithms are usually evaluated on a single dataset. Although training and testing are performed on different subsets of the dataset, models built on one study show limited capability to generalize to other studies. While database bias has been recognized as a serious problem in the computer vision community, it has remained largely unnoticed in medical imaging research. Transfer learning thus remains confined to the re-use of feature representations requiring re-training on the new dataset. As a result, machine learning models do not generalize even when trained on imaging datasets that were captured to study the same variable of interest. The ability to transfer knowledge gleaned from one study to another, without the need for re-training, if possible, would provide reassurance that the models are learning knowledge fundamental to the problem under study instead of latching onto the idiosyncracies of a dataset. In this paper, we situate the problem of dataset bias in the context of medical imaging studies. We show empirical evidence that such a problem exists in medical datasets. We then present a framework to unlearn study membership as a means to handle the problem of database bias. Our main idea is to take the data from the original feature space to an intermediate space where the data points are indistinguishable in terms of which study they come from, while maintaining the recognition capability with respect to the variable of interest. This will promote models which learn the more general properties of the etiology under study instead of aligning to dataset-specific peculiarities. Essentially, our proposed model learns to unlearn the dataset bias.
37.Deep Learning Architect: Classification for Architectural Design through the Eye of Artificial Intelligence pdf
This paper applies state-of-the-art techniques in deep learning and computer vision to measure visual similarities between architectural designs by different architects. Using a dataset consisting of web scraped images and an original collection of images of architectural works, we first train a deep convolutional neural network (DCNN) model capable of achieving 73% accuracy in classifying works belonging to 34 different architects. Through examining the weights in the trained DCNN model, we are able to quantitatively measure the visual similarities between architects that are implicitly learned by our model. Using this measure, we cluster architects that are identified to be similar and compare our findings to conventional classification made by architectural historians and theorists. Our clustering of architectural designs remarkably corroborates conventional views in architectural history, and the learned architectural features also coheres with the traditional understanding of architectural designs.
38.FineFool: Fine Object Contour Attack via Attention pdf
Machine learning models have been shown vulnerable to adversarial attacks launched by adversarial examples which are carefully crafted by attacker to defeat classifiers. Deep learning models cannot escape the attack either. Most of adversarial attack methods are focused on success rate or perturbations size, while we are more interested in the relationship between adversarial perturbation and the image itself. In this paper, we put forward a novel adversarial attack based on contour, named FineFool. Finefool not only has better attack performance compared with other state-of-art white-box attacks in aspect of higher attack success rate and smaller perturbation, but also capable of visualization the optimal adversarial perturbation via attention on object contour. To the best of our knowledge, Finefool is for the first time combines the critical feature of the original clean image with the optimal perturbations in a visible manner. Inspired by the correlations between adversarial perturbations and object contour, slighter perturbations is produced via focusing on object contour features, which is more imperceptible and difficult to be defended, especially network add-on defense methods with the trade-off between perturbations filtering and contour feature loss. Compared with existing state-of-art attacks, extensive experiments are conducted to show that Finefool is capable of efficient attack against defensive deep models.
39.Multiview Based 3D Scene Understanding On Partial Point Sets pdf
Deep learning within the context of point clouds has gained much research interest in recent years mostly due to the promising results that have been achieved on a number of challenging benchmarks, such as 3D shape recognition and scene semantic segmentation. In many realistic settings however, snapshots of the environment are often taken from a single view, which only contains a partial set of the scene due to the field of view restriction of commodity cameras. 3D scene semantic understanding on partial point clouds is considered as a challenging task. In this work, we propose a processing approach for 3D point cloud data based on a multiview representation of the existing 360° point clouds. By fusing the original 360° point clouds and their corresponding 3D multiview representations as input data, a neural network is able to recognize partial point sets while improving the general performance on complete point sets, resulting in an overall increase of 31.9% and 4.3% in segmentation accuracy for partial and complete scene semantic understanding, respectively. This method can also be applied in a wider 3D recognition context such as 3D part segmentation.
40.A Graph-CNN for 3D Point Cloud Classification pdf
Graph convolutional neural networks (Graph-CNNs) extend traditional CNNs to handle data that is supported on a graph. Major challenges when working with data on graphs are that the support set (the vertices of the graph) do not typically have a natural ordering, and in general, the topology of the graph is not regular (i.e., vertices do not all have the same number of neighbors). Thus, Graph-CNNs have huge potential to deal with 3D point cloud data which has been obtained from sampling a manifold. In this paper, we develop a Graph-CNN for classifying 3D point cloud data, called PointGCN. The architecture combines localized graph convolutions with two types of graph downsampling operations (also known as pooling). By the effective exploration of the point cloud local structure using the Graph-CNN, the proposed architecture achieves competitive performance on the 3D object classification benchmark ModelNet, and our architecture is more stable than competing schemes.
41.GANtruth - an unpaired image-to-image translation method for driving scenarios pdf
Synthetic image translation has significant potentials in autonomous transportation systems. That is due to the expense of data collection and annotation as well as the unmanageable diversity of real-words situations. The main issue with unpaired image-to-image translation is the ill-posed nature of the problem. In this work, we propose a novel method for constraining the output space of unpaired image-to-image translation. We make the assumption that the environment of the source domain is known (e.g. synthetically generated), and we propose to explicitly enforce preservation of the ground-truth labels on the translated images.
We experiment on preserving ground-truth information such as semantic segmentation, disparity, and instance segmentation. We show significant evidence that our method achieves improved performance over the state-of-the-art model of UNIT for translating images from SYNTHIA to Cityscapes. The generated images are perceived as more realistic in human surveys and outperforms UNIT when used in a domain adaptation scenario for semantic segmentation.
42.Assigning a Grade: Accurate Measurement of RoadQuality Using Satellite Imagery pdf
Roads are critically important infrastructure to societal and economic development, with huge investments made by governments every year. However, methods for monitoring those investments tend to be time-consuming, laborious, and expensive, placing them out of reach for many developing regions. In this work, we develop a model for monitoring the quality of road infrastructure using satellite imagery. For this task, we harness two trends: the increasing availability of high-resolution, often-updated satellite imagery, and the enormous improvement in speed and accuracy of convolutional neural network-based methods for performing computer vision tasks. We employ a unique dataset of road quality information on 7000km of roads in Kenya combined with 50cm resolution satellite imagery. We create models for a binary classification task as well as a comprehensive 5-category classification task, with accuracy scores of 88 and 73 percent respectively. We also provide evidence of the robustness of our methods with challenging held-out scenarios, though we note some improvement is still required for confident analysis of a never before seen road. We believe these results are well-positioned to have substantial impact on a broad set of transport applications.
43.General-to-Detailed GAN for Infrequent Class Medical Images pdf
Deep learning has significant potential for medical imaging. However, since the incident rate of each disease varies widely, the frequency of classes in a medical image dataset is imbalanced, leading to poor accuracy for such infrequent classes. One possible solution is data augmentation of infrequent classes using synthesized images created by Generative Adversarial Networks (GANs), but conventional GANs also require certain amount of images to learn. To overcome this limitation, here we propose General-to-detailed GAN (GDGAN), serially connected two GANs, one for general labels and the other for detailed labels. GDGAN produced diverse medical images, and the network trained with an augmented dataset outperformed other networks using existing methods with respect to Area-Under-Curve (AUC) of Receiver Operating Characteristic (ROC) curve.
44.Training for 'Unstable' CNN Accelerator:A Case Study on FPGA pdf
With the great advancements of convolution neural networks(CNN), CNN accelerators are increasingly developed and deployed in the major computing systems.To make use of the CNN accelerators, CNN models are trained via the off-line training systems such as Caffe, Pytorch and Tensorflow on multi-core CPUs and GPUs first and then compiled to the target accelerators. Although the two-step process seems to be natural and has been widely applied, it assumes that the accelerators' behavior can be fully modeled on CPUs and GPUs. This does not hold true and the behavior of the CNN accelerators is un-deterministic when the circuit works at 'unstable' mode when it is overclocked or is affected by the environment like fault-prone aerospace. The exact behaviors of the accelerators are determined by both the chip fabrication and the working environment or status. In this case, applying the conventional off-line training result to the accelerators directly may lead to considerable accuracy loss.
To address this problem, we propose to train for the 'unstable' CNN accelerator and have the 'un-determined behavior' learned together with the data in the same framework. Basically, it starts from the off-line trained model and then integrates the uncertain circuit behaviors into the CNN models through additional accelerator-specific training. The fine-tuned training makes the CNN models less sensitive to the circuit uncertainty. We apply the design method to both an overclocked CNN accelerator and a faulty accelerator. According to our experiments on a subset of ImageNet, the accelerator-specific training can improve the top 5 accuracy up to 3.4% and 2.4% on average when the CNN accelerator is at extreme overclocking. When the accelerator is exposed to a faulty environment, the top 5 accuracy improves up to 6.8% and 4.28% on average under the most severe fault injection.
45.Learning Saliency Maps for Adversarial Point-Cloud Generation pdf
3D point-cloud recognition with deep neural network (DNN) has received remarkable progress on obtaining both high-accuracy recognition and robustness to random point missing (or dropping). However, the robustness of DNNs to maliciously-manipulated point missing is still unclear. In this paper, we show that point-missing can be a critical security concern by proposing a {\em malicious point-dropping method} to generate adversarial point clouds to fool DNNs. Our method is based on learning a saliency map for a whole point cloud, which assigns each point a score reflecting its contribution to the model-recognition loss, i.e., the difference between the losses with and without the specific point respectively. The saliency map is learnt by approximating the nondifferentiable point-dropping process with a differentiable procedure of shifting points towards the cloud center. In this way, the loss difference, i.e., the saliency score for each point in the map, can be measured by the corresponding gradient of the loss w.r.t the point under the spherical coordinates. Based on the learned saliency map, maliciously point-dropping attack can be achieved by dropping points with the highest scores, leading to significant increase of model loss and thus inferior classification performance. Extensive evaluations on several state-of-the-art point-cloud recognition models, including PointNet, PointNet++ and DGCNN, demonstrate the efficacy and generality of our proposed saliency-map-based point-dropping scheme. Code for experiments is released on \url{this https URL}.
46.Deep Bayesian Uncertainty Estimation for Adaptation and Self-Annotation of Food Packaging Images pdf
Food packaging labels provide important information for public health, such as allergens and use-by dates. Off-the-shelf Optical Character Verification (OCV) systems are good solutions for automating food label quality assessments, but are known to under perform on complex data. This paper proposes a Deep Learning based system that can identify inadequate images for OCV, due to their poor label quality, by employing state-of-the-art Convolutional Neural Network (CNN) architectures, and practical Bayesian inference techniques for automatic self-annotation. We propose a practical domain adaptation procedure based on k-means clustering of CNN latent variables, followed by a k-Nearest Neighbour classification for handling high label variability between different dataset distributions. Moreover, Supervised Learning has proven useful in such systems but manual annotation of large amounts of data is usually required. This is practically intractable in most real world problems due to time/labour constraints. In an attempt to address this issue, we introduce a self-annotating prediction model based on Self-Training of a Bayesian CNN, that leverages modern variational inference methods of deep models. In this context, we propose a new inverse uncertainty weighting technique that encourages the Self-Training model to learn from more informative data over time, potentially preventing it from becoming lazy by only selecting easy examples to learn from. An experimental study is presented illustrating the superior performance of the proposed approach over standard Self-Training, and highlighting the importance of predictive uncertainty estimates in safety-critical domains.
47.A Pixel-Based Framework for Data-Driven Clothing pdf
With the aim of creating virtual cloth deformations more similar to real world clothing, we propose a new computational framework that recasts three dimensional cloth deformation as an RGB image in a two dimensional pattern space. Then a three dimensional animation of cloth is equivalent to a sequence of two dimensional RGB images, which in turn are driven/choreographed via animation parameters such as joint angles. This allows us to leverage popular CNNs to learn cloth deformations in image space. The two dimensional cloth pixels are extended into the real world via standard body skinning techniques, after which the RGB values are interpreted as texture offsets and displacement maps. Notably, we illustrate that our approach does not require accurate unclothed body shapes or robust skinning techniques. Additionally, we discuss how standard image based techniques such as image partitioning for higher resolution, GANs for merging partitioned image regions back together, etc., can readily be incorporated into our framework.
48.Learning to Sample pdf
Processing large point clouds is a challenging task. Therefore, the data is often sampled to a size that can be processed more easily. The question is how to sample the data? A popular sampling technique is Farthest Point Sampling (FPS). However, FPS is agnostic to a downstream application (classification, retrieval, etc.). The underlying assumption seems to be that minimizing the farthest point distance, as done by FPS, is a good proxy to other objective functions.
We show that it is better to learn how to sample. To do that, we propose a deep network to simplify 3D point clouds. The network, termed S-NET, takes a point cloud and produces a smaller point cloud that is optimized for a particular task. The simplified point cloud is not guaranteed to be a subset of the original point cloud. Therefore, we match it to a subset of the original points in a post-processing step. We contrast our approach with FPS by experimenting on two standard data sets and show significantly better results for a variety of applications.
49.Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling pdf
The unconditional generation of high fidelity images is a longstanding benchmark for testing the performance of image decoders. Autoregressive image models have been able to generate small images unconditionally, but the extension of these methods to large images where fidelity can be more readily assessed has remained an open problem. Among the major challenges are the capacity to encode the vast previous context and the sheer difficulty of learning a distribution that preserves both global semantic coherence and exactness of detail. To address the former challenge, we propose the Subscale Pixel Network (SPN), a conditional decoder architecture that generates an image as a sequence of sub-images of equal size. The SPN compactly captures image-wide spatial dependencies and requires a fraction of the memory and the computation required by other fully autoregressive models. To address the latter challenge, we propose to use Multidimensional Upscaling to grow an image in both size and depth via intermediate stages utilising distinct SPNs. We evaluate SPNs on the unconditional generation of CelebAHQ of size 256 and of ImageNet from size 32 to 256. We achieve state-of-the-art likelihood results in multiple settings, set up new benchmark results in previously unexplored settings and are able to generate very high fidelity large scale samples on the basis of both datasets.
50.Training Competitive Binary Neural Networks from Scratch pdf
Convolutional neural networks have achieved astonishing results in different application areas. Various methods that allow us to use these models on mobile and embedded devices have been proposed. Especially binary neural networks are a promising approach for devices with low computational power. However, training accurate binary models from scratch remains a challenge. Previous work often uses prior knowledge from full-precision models and complex training strategies. In our work, we focus on increasing the performance of binary neural networks without such prior knowledge and a much simpler training strategy. In our experiments we show that we are able to achieve state-of-the-art results on standard benchmark datasets. Further, to the best of our knowledge, we are the first to successfully adopt a network architecture with dense connections for binary networks, which lets us improve the state-of-the-art even further.
51.Video Synthesis from a Single Image and Motion Stroke pdf
In this paper, we propose a new method to automatically generate a video sequence from a single image and a user provided motion stroke. Generating a video sequence based on a single input image has many applications in visual content creation, but it is tedious and time-consuming to produce even for experienced artists. Automatic methods have been proposed to address this issue, but most existing video prediction approaches require multiple input frames. In addition, generated sequences have limited variety since the output is mostly determined by the input frames, without allowing the user to provide additional constraints on the result. In our technique, users can control the generated animation using a sketch stroke on a single input image. We train our system such that the trajectory of the animated object follows the stroke, which makes it both more flexible and more controllable. From a single image, users can generate a variety of video sequences corresponding to different sketch inputs. Our method is the first system that, given a single frame and a motion stroke, can generate animations by recurrently generating videos frame by frame. An important benefit of the recurrent nature of our architecture is that it facilitates the synthesis of an arbitrary number of generated frames. Our architecture uses an autoencoder and a generative adversarial network (GAN) to generate sharp texture images, and we use another GAN to guarantee that transitions between frames are realistic and smooth. We demonstrate the effectiveness of our approach on the MNIST, KTH, and Human 3.6M datasets.
52.Regularized Ensembles and Transferability in Adversarial Learning pdf
Despite the considerable success of convolutional neural networks in a broad array of domains, recent research has shown these to be vulnerable to small adversarial perturbations, commonly known as adversarial examples. Moreover, such examples have shown to be remarkably portable, or transferable, from one model to another, enabling highly successful black-box attacks. We explore this issue of transferability and robustness from two dimensions: first, considering the impact of conventional
$l_p$ regularization as well as replacing the top layer with a linear support vector machine (SVM), and second, the value of combining regularized models into an ensemble. We show that models trained with different regularizers present barriers to transferability, as does partial information about the models comprising the ensemble.