Driver attention prediction has gained significant attention recently due to its role in developing advanced driver assistance systems (ADAS) and intelligent vehicles. The emergence of video foundation models (VFMs) has opened up new possibilities for improving video understanding tasks like video saliency prediction (VSP). However, these large models are often not cost-effective for ADAS and intelligent vehicles due to their size and resource demands. To address this, we present an early effort to use knowledge distillation for predicting driver visual attention, employing the first VFM-based VSP model, SalFoM, as the teacher network. Given that driver attention prediction datasets are smaller than those used for large models, fine-tuning such models is challenging due to their high parameter count. To overcome this, we designed a VFM-based driver attention prediction network with fewer parameters than the teacher network. Experimental results show our model’s effectiveness on benchmark datasets.

With vast amounts of video content being uploaded to the Internet every minute, video summarization becomes critical for efficient browsing, searching, and indexing of visual content. Nonetheless, the spread of social and egocentric cameras creates an abundance of sparse scenarios captured by several devices, and ultimately required to be jointly summarized. In this paper, we discuss the problem of summarizing videos recorded independently by several dynamic cameras that intermittently share the field of view. We present a robust framework that (a) identifies a diverse set of important events among moving cameras that often are not capturing the same scene, and (b) selects the most representative view(s) at each event to be included in a universal summary. Due to the lack of an applicable alternative, we collected a new multi-view egocentric dataset, Multi-Ego. Our dataset is recorded simultaneously by three cameras, covering a wide variety of real-life scenarios. The footage is annotated by multiple individuals under various summarization configurations, with a consensus analysis ensuring a reliable ground truth. We conduct extensive experiments on the compiled dataset in addition to three other standard benchmarks that show the robustness and the advantage of our approach in both supervised and unsupervised settings. Additionally, we show that our approach learns collectively from data of varied number-of-views and orthogonal to other summarization methods, deeming it scalable and generic.

CNN-based salient object detection (SOD) methods achieve impressive performance. However, the way semantic information is encoded in them and whether they are category-agnostic is less explored. One major obstacle in studying these questions is the fact that SOD models are built on top of the ImageNet pre-trained backbones which may cause information leakage and feature redundancy. To remedy this, here we first propose an extremely light-weight holistic model tied to the SOD task that can be freed from classification backbones and trained from scratch, and then employ it to study the semantics of SOD models. With the holistic network and representation redundancy reduction by a novel dynamic weight decay scheme, our model has only 100K parameters, ∼ 0.2% of parameters of large models, and performs on par with SOTA on popular SOD benchmarks. Using CSNet, we find that a) SOD and classification methods use different mechanisms, b) SOD models are category insensitive, c) ImageNet pre-training is not necessary for SOD training, and d) SOD models require far fewer parameters than the classification models. The source code is publicly available at https://mmcheng.net/sod100k/.

Deep neural networks are vulnerable to adversarial attacks. More importantly, some adversarial examples crafted against an ensemble of source models transfer to other target models and, thus, pose a security threat to black-box applications (when attackers have no access to the target models). Current transfer-based ensemble attacks, however, only consider a limited number of source models to craft an adversarial example and, thus, obtain poor transferability. Besides, recent query-based black-box attacks, which require numerous queries to the target model, not only come under suspicion by the target model but also cause expensive query cost. In this article, we propose a novel transfer-based black-box attack, dubbed serial-minigroup-ensemble-attack (SMGEA). Concretely, SMGEA first divides a large number of pretrained white-box source models into several "minigroups." For each minigroup, we design three new ensemble strategies to improve the intragroup transferability. Moreover, we propose a new algorithm that recursively accumulates the "long-term" gradient memories of the previous minigroup to the subsequent minigroup. This way, the learned adversarial information can be preserved, and the intergroup transferability can be improved. Experiments indicate that SMGEA not only achieves state-of-the-art black-box attack ability over several data sets but also deceives two online black-box saliency prediction systems in real world, i.e., DeepGaze-II (https://deepgaze.bethgelab.org/) and SALICON (http://salicon.net/demo/). Finally, we contribute a new code repository to promote research on adversarial attack and defense over ubiquitous pixel-to-pixel computer vision tasks. We share our code together with the pretrained substitute model zoo at https://github.com/CZHQuality/AAA-Pix2pix.

Visual saliency models have enjoyed a big leap in performance in recent years, thanks to advances in deep learning and large scale annotated data. Despite enormous effort and huge breakthroughs, however, models still fall short in reaching human-level accuracy. In this work, I explore the landscape of the field emphasizing on new deep saliency models, benchmarks, and datasets. A large number of image and video saliency models are reviewed and compared over two image benchmarks and two large scale video datasets. Further, I identify factors that contribute to the gap between models and humans and discuss the remaining issues that need to be addressed to build the next generation of more powerful saliency models. Some specific questions that are addressed include: in what ways current models fail, how to remedy them, what can be learned from cognitive studies of attention, how explicit saliency judgments relate to fixations, how to conduct fair model comparison, and what are the emerging applications of saliency models.

Predicting where people look in static scenes, a.k.a visual saliency, has received significant research interest recently. However, relatively less effort has been spent in understanding and modeling visual attention over dynamic scenes. This work makes three contributions to video saliency research. First, we introduce a new benchmark, called DHF1K (Dynamic Human Fixation 1K), for predicting fixations during dynamic scene free-viewing, which is a long-time need in this field. DHF1K consists of 1K high-quality elaborately-selected video sequences annotated by 17 observers using an eye tracker device. The videos span a wide range of scenes, motions, object types and backgrounds. Second, we propose a novel video saliency model, called ACLNet (Attentive CNN-LSTM Network), that augments the CNN-LSTM architecture with a supervised attention mechanism to enable fast end-to-end saliency learning. The attention mechanism explicitly encodes static saliency information, thus allowing LSTM to focus on learning a more flexible temporal saliency representation across successive frames. Such a design fully leverages existing large-scale static fixation datasets, avoids overfitting, and significantly improves training efficiency and testing performance. Third, we perform an extensive evaluation of the state-of-the-art saliency models on three datasets : DHF1K, Hollywood-2, and UCF sports. An attribute-based analysis of previous saliency models and cross-dataset generalization are also presented. Experimental results over more than 1.2K testing videos containing 400K frames demonstrate that ACLNet outperforms other contenders and has a fast processing speed (40 fps using a single GPU). Our code and all the results are available at https://github.com/wenguanwang/DHF1K.

Saliency detection is an effective front-end process to many security-related tasks, e.g. automatic drive and tracking. Adversarial attack serves as an efficient surrogate to evaluate the robustness of deep saliency models before they are deployed in real world. However, most of current adversarial attacks exploit the gradients spanning the entire image space to craft adversarial examples, ignoring the fact that natural images are high-dimensional and spatially over-redundant, thus causing expensive attack cost and poor perceptibility. To circumvent these issues, this paper builds an efficient bridge between the accessible partially-white-box source models and the unknown black-box target models. The proposed method includes two steps: 1) We design a new partially-white-box attack, which defines the cost function in the compact hidden space to punish a fraction of feature activations corresponding to the salient regions, instead of punishing every pixel spanning the entire dense output space. This partially-white-box attack reduces the redundancy of the adversarial perturbation. 2) We exploit the non-redundant perturbations from some source models as the prior cues, and use an iterative zeroth-order optimizer to compute the directional derivatives along the non-redundant prior directions, in order to estimate the actual gradient of the black-box target model. The non-redundant priors boost the update of some "critical" pixels locating at non-zero coordinates of the prior cues, while keeping other redundant pixels locating at the zero coordinates unaffected. Our method achieves the best tradeoff between attack ability and perturbation redundancy. Finally, we conduct a comprehensive experiment to test the robustness of 18 state-of-the-art deep saliency models against 16 malicious attacks, under both of white-box and black-box settings, which contributes a new robustness benchmark to the saliency community for the first time.

Multi-level feature fusion is a fundamental topic in computer vision. It has been exploited to detect, segment and classify objects at various scales. When multi-level features meet multi-modal cues, the optimal feature aggregation and multi-modal learning strategy become a hot potato. In this paper, we leverage the inherent multi-modal and multi-level nature of RGB-D salient object detection to devise a novel Bifurcated Backbone Strategy Network (BBS-Net). Our architecture, is simple, efficient, and backbone-independent. In particular, first, we propose to regroup the multi-level features into teacher and student features using a bifurcated backbone strategy (BBS). Second, we introduce a depth-enhanced module (DEM) to excavate informative depth cues from the channel and spatial views. Then, RGB and depth modalities are fused in a complementary way. Extensive experiments show that BBS-Net significantly outperforms 18 state-of-the-art (SOTA) models on eight challenging datasets under five evaluation measures, demonstrating the superiority of our approach (~4% improvement in S-measure vs . the top-ranked model: DMRA). In addition, we provide a comprehensive analysis on the generalization ability of different RGB-D datasets and provide a powerful training set for future research. The complete algorithm, benchmark results, and post-processing toolbox are publicly available at https://github.com/zyjwuyan/BBS-Net.

Saliency prediction in traditional images and videos has drawn extensive research interests in recent years. Few works have been proposed for saliency prediction over 360° videos. They focus on directly predicting fixations over the whole panorama. When viewing 360° videos, a person can only observe the content in her viewport, which means that only a fraction of the 360° scene can be seen at any given time. In this paper, we study human attention over viewport of 360° videos and propose a novel visual saliency model, dubbed viewport saliency, to predict fixations over 360° videos. Two contributions are introduced. First, we find that where people look is affected by the content and location of the viewport in 360° video. We study this over 200+ 360° videos viewed by 30+ subjects over two recent benchmark databases. Second, we propose a Multi-Task Deep Neural Network (MT-DNN) method for Viewport Saliency (VS) prediction in 360° video, which considers the input content and location of the viewport. Extensive experiments and analyses show that our method outperforms other state-of-the-art methods in this task. In particular, over the two recent 360° video databases, our MT-DNN raises the average CC score by 0.149 and 0.205, compared to SalGAN and DeepVS methods, respectively.

Deep neural networks are vulnerable to adversarial attacks. More importantly, some adversarial examples crafted against an ensemble of pre-trained source models can transfer to other new target models, thus pose a security threat to black-box applications (when the attackers have no access to the target models). Despite adopting diverse architectures and parameters, source and target models often share similar decision boundaries. Therefore, if an adversary is capable of fooling several source models concurrently, it can potentially capture intrinsic transferable adversarial information that may allow it to fool a broad class of other black-box target models. Current ensemble attacks, however, only consider a limited number of source models to craft an adversary, and obtain poor transferability. In this paper, we propose a novel black-box attack, dubbed Serial-Mini-Batch-Ensemble-Attack (SMBEA). SMBEA divides a large number of pre-trained source models into several mini-batches. For each single batch, we design 3 new ensemble strategies to improve the intra-batch transferability. Besides, we propose a new algorithm that recursively accumulates the “long-term” gradient memories of the previous batch to the following batch. This way, the learned adversarial information can be preserved and the inter-batch transferability can be improved. Experiments indicate that our method outperforms state-of-the-art ensemble attacks over multiple pixel-to-pixel vision tasks including image translation and salient region prediction. Our method successfully fools two online black-box saliency prediction systems including DeepGaze-II (Kummerer 2017) and SALICON (Huang et al. 2017). Finally, we also contribute a new repository to promote the research on adversarial attack and defense over pixel-to-pixel tasks: https://github.com/CZHQuality/AAA-Pix2pix.

Previous research in visual saliency has been focused on two major types of models namely fixation prediction and salient object detection. The relationship between the two, however, has been less explored. In this work, we propose to employ the former model type to identify salient objects. We build a novel Attentive Saliency Network (ASNet)11.Available at: https://github.com/wenguanwang/ASNet. that learns to detect salient objects from fixations. The fixation map, derived at the upper network layers, mimics human visual attention mechanisms and captures a high-level understanding of the scene from a global view. Salient object detection is then viewed as fine-grained object-level saliency segmentation and is progressively optimized with the guidance of the fixation map in a top-down manner. ASNet is based on a hierarchy of convLSTMs that offers an efficient recurrent mechanism to sequentially refine the saliency features over multiple steps. Several loss functions, derived from existing saliency evaluation metrics, are incorporated to further boost the performance. Extensive experiments on several challenging datasets show that our ASNet outperforms existing methods and is capable of generating accurate segmentation maps with the help of the computed fixation prior. Our work offers a deeper insight into the mechanisms of attention and narrows the gap between salient object detection and fixation prediction.

Humans are able to track multiple objects at any given time in their daily activities-for example, we can drive a car while monitoring obstacles, pedestrians, and other vehicles. Several past studies have examined how humans track targets simultaneously and what underlying behavioral and neural mechanisms they use. At the same time, computer-vision researchers have proposed different algorithms to track multiple targets automatically. These algorithms are useful for video surveillance, team-sport analysis, video analysis, video summarization, and human-computer interaction. Although there are several efficient biologically inspired algorithms in artificial intelligence, the human multiple-target tracking (MTT) ability is rarely imitated in computer-vision algorithms. In this paper, we review MTT studies in neuroscience and biologically inspired MTT methods in computer vision and discuss the ways in which they can be seen as complementary.

Data size is the bottleneck for developing deep saliency models, because collecting eye-movement data is very time-consuming and expensive. Most of current studies on human attention and saliency modeling have used high-quality stereotype stimuli. In real world, however, captured images undergo various types of transformations. Can we use these transformations to augment existing saliency datasets? Here, we first create a novel saliency dataset including fixations of 10 observers over 1900 images degraded by 19 types of transformations. Second, by analyzing eye movements, we find that observers look at different locations over transformed versus original images. Third, we utilize the new data over transformed images, called data augmentation transformation (DAT), to train deep saliency models. We find that label-preserving DATs with negligible impact on human gaze boost saliency prediction, whereas some other DATs that severely impact human gaze degrade the performance. These label-preserving valid augmentation transformations provide a solution to enlarge existing saliency datasets. Finally, we introduce a novel saliency model based on generative adversarial networks (dubbed GazeGAN). A modified U-Net is utilized as the generator of the GazeGAN, which combines classic "skip connection" with a novel "center-surround connection" (CSC) module. Our proposed CSC module mitigates trivial artifacts while emphasizing semantic salient regions, and increases model nonlinearity, thus demonstrating better robustness against transformations. Extensive experiments and comparisons indicate that GazeGAN achieves state-of-the-art performance over multiple datasets. We also provide a comprehensive comparison of 22 saliency models on various transformed scenes, which contributes a new robustness benchmark to saliency community. Our code and dataset are available at.

Deep convolutional neural networks (CNNs) have been successfully applied to a wide variety of problems in computer vision, including salient object detection. To accurately detect and segment salient objects, it is necessary to extract and combine high-level semantic features with low-level fine details simultaneously. This is challenging for CNNs because repeated subsampling operations such as pooling and convolution lead to a significant decrease in the feature resolution, which results in the loss of spatial details and finer structures. Therefore, we propose augmenting feedforward neural networks by using the multistage refinement mechanism. In the first stage, a master net is built to generate a coarse prediction map in which most detailed structures are missing. In the following stages, the refinement net with layerwise recurrent connections to the master net is equipped to progressively combine local context information across stages to refine the preceding saliency maps in a stagewise manner. Furthermore, the pyramid pooling module and channel attention module are applied to aggregate different-region-based global contexts. Extensive evaluations over six benchmark datasets show that the proposed method performs favorably against the state-of-the-art approaches.

Recently, data-driven deep saliency models have achieved high performance and have outperformed classical saliency models, as demonstrated by results on datasets such as the MIT300 and SALICON. Yet, there remains a large gap between the performance of these models and the inter-human baseline. Some outstanding questions include what have these models learned, how and where they fail, and how they can be improved. This article attempts to answer these questions by analyzing the representations learned by individual neurons located at the intermediate layers of deep saliency models. To this end, we follow the steps of existing deep saliency models, that is borrowing a pre-trained model of object recognition to encode the visual features and learning a decoder to infer the saliency. We consider two cases when the encoder is used as a fixed feature extractor and when it is fine-tuned, and compare the inner representations of the network. To study how the learned representations depend on the task, we fine-tune the same network using the same image set but for two different tasks: saliency prediction versus scene classification. Our analyses reveal that: 1) some visual regions (e.g. head, text, symbol, vehicle) are already encoded within various layers of the network pre-trained for object recognition, 2) using modern datasets, we find that fine-tuning pre-trained models for saliency prediction makes them favor some categories (e.g. head) over some others (e.g. text), 3) although deep models of saliency outperform classical models on natural images, the converse is true for synthetic stimuli (e.g. pop-out search arrays), an evidence of significant difference between human and data-driven saliency models, and 4) we confirm that, after-fine tuning, the change in inner-representations is mostly due to the task and not the domain shift in the data.

This paper presents a new method for detecting salient objects in images using convolutional neural networks (CNNs). The proposed network, named PAGE-Net, offers two key contributions. The first is the exploitation of an essential pyramid attention structure for salient object detection. This enables the network to concentrate more on salient regions while considering multi-scale saliency information. Such a stacked attention design provides a powerful tool to efficiently improve the representation ability of the corresponding network layer with an enlarged receptive field. The second contribution lies in the emphasis on the importance of salient edges. Salient edge information offers a strong cue to better segment salient objects and refine object boundaries. To this end, our model is equipped with a salient edge detection module, which is learned for precise salient boundary estimation. This encourages better edge-preserving salient object segmentation. Exhaustive experiments confirm that the proposed pyramid attention and salient edges are effective for salient object detection. We show that our deep saliency model outperforms state-of-the-art approaches for several benchmarks with a fast processing speed (25fps on one GPU).

In this work, we present a novel dataset consisting of eye movements and verbal descriptions recorded synchronously over images. Using this data, we study the differences in human attention during free-viewing and image captioning tasks. We look into the relationship between human atten- tion and language constructs during perception and sen- tence articulation. We also analyse attention deployment mechanisms in the top-down soft attention approach that is argued to mimic human attention in captioning tasks, and investigate whether visual saliency can help image caption- ing. Our study reveals that (1) human attention behaviour differs in free-viewing and image description tasks. Hu- mans tend to fixate on a greater variety of regions under the latter task, (2) there is a strong relationship between de- scribed objects and attended objects (97% of the described objects are being attended), (3) a convolutional neural net- work as feature encoder accounts for human-attended re- gions during image captioning to a great extent (around 78%), (4) soft-attention mechanism differs from human at- tention, both spatially and temporally, and there is low correlation between caption scores and attention consis- tency scores. These indicate a large gap between humans and machines in regards to top-down attention, and (5) by integrating the soft attention model with image saliency, we can significantly improve the model's performance on Flickr30k and MSCOCO benchmarks. The dataset can be found at: https://github.com/SenHe/ Human-Attention-in-Image-Captioning.

To automatically produce a brief yet expressive summary of a long video, an automatic algorithm should start by resembling the human process of summary generation. Prior work proposed supervised and unsupervised algorithms to train models for learning the underlying behavior of humans by increasing modeling complexity or craft-designing better heuristics to simulate human summary generation process. In this work, we take a different approach by analyzing a major cue that humans exploit for summary generation; the nature and intensity of actions. We empirically observed that a frame is more likely to be included in human-generated summaries if it contains a substantial amount of deliberate motion performed by an agent, which is referred to as actionness. Therefore, we hypothesize that learning to automatically generate summaries involves an implicit knowledge of actionness estimation and ranking. We validate our hypothesis by running a user study that explores the correlation between human-generated summaries and actionness ranks. We also run a consensus and behavioral analysis between human subjects to ensure reliable and consistent results. The analysis exhibits a considerable degree of agreement among subjects within obtained data and verifying our initial hypothesis. Based on the study findings, we develop a method to incorporate actionness data to explicitly regulate a learning algorithm that is trained for summary generation. We assess the performance of our approach on 4 summarization benchmark datasets, and demonstrate an evident advantage compared to state-of-the-art summarization methods.

This work addresses sky segmentation, the task of determining sky and non-sky pixels in images, and improving upon existing state-of-the-art models. Outdoor scene parsing models are often trained on ideal datasets and produce high-quality results. However, this leads to inferior performance when applied to real-world images. The quality of scene parsing, particularly sky segmentation, decreases in night-time images, images involving varying weather conditions, and scene changes due to seasonal weather. We address these challenges using the RefineNet model in conjunction with two datasets: SkyFinder, and a subset of the SUN database containing sky regions (SUN-sky, henceforth). We achieve an improvement of 10-15% in the average MCR compared to prior methods using the SkyFinder dataset, and nearly 36% improvement from an off-the-shelf model in terms of average mIOU score. Employing fully connected conditional random fields as a post processing method demonstrates further enhancement of our results. Furthermore, by analyzing models over images with respect to two aspects, time of day and weather conditions, we find that when facing the same challenges as prior methods, our trained models significantly outperform them.

This paper digs deeper into factors that influence egocentric gaze. Instead of training deep models for this purpose in a blind manner, we propose to inspect factors that contribute to gaze guidance during daily tasks. Bottom-up saliency and optical flow are assessed versus strong spatial prior baselines. Task-specific cues such as vanishing point, manipulation point, and hand regions are analyzed as representatives of top-down information. We also look into the contribution of these factors by investigating a simple recurrent neural model for ego-centric gaze prediction. First, deep features are extracted for all input video frames. Then, a gated recurrent unit is employed to integrate information over time and to predict the next fixation. We also propose an integrated model that combines the recurrent model with several top-down and bottom-up cues. Extensive experiments over multiple datasets reveal that (1) spatial biases are strong in egocentric videos, (2) bottom-up saliency models perform poorly in predicting gaze and underperform spatial biases, (3) deep features perform better compared to traditional features, (4) as opposed to hand regions, the manipulation point is a strong influential cue for gaze prediction, (5) combining the proposed recurrent model with bottom-up cues, vanishing points and, in particular, manipulation point results in the best gaze prediction accuracy over egocentric videos, (6) the knowledge transfer works best for cases where the tasks or sequences are similar, and (7) task and activity recognition can benefit from gaze prediction. Our findings suggest that (1) there should be more emphasis on hand-object interaction and (2) the egocentric vision community should consider larger datasets including diverse stimuli and more subjects.

Generative models , in particular generative adversarial networks (GANs), have gained significant attention in recent years. A number of GAN variants have been proposed and have been utilized in many applications. Despite large strides in terms of theoretical progress, evaluating and comparing GANs remains a daunting task. While several measures have been introduced, as of yet, there is no consensus as to which measure best captures strengths and limitations of models and should be used for fair model comparison. As in other areas of computer vision and machine learning , it is critical to settle on one or few good measures to steer the progress in this field. In this paper, I review and critically discuss more than 24 quantitative and 5 qualitative measures for evaluating generative models with a particular emphasis on GAN-derived models. I also provide a set of 7 desiderata followed by an evaluation of whether a given measure or a family of measures is compatible with them.

Detecting and segmenting salient objects from natural scenes, often referred to as salient object detection, has attracted great interest in computer vision. While many models have been proposed and several applications have emerged, a deep understanding of achievements and issues remains lacking. We aim to provide a comprehensive review of recent progress in salient object detection and situate this field among other closely related areas such as generic scene segmentation, object proposal generation, and saliency for fixation prediction. Covering 228 publications, we survey i) roots, key concepts, and tasks, ii) core techniques and main modeling trends, and iii) datasets and evaluation metrics for salient object detection. We also discuss open problems such as evaluation metrics and dataset bias in model performance, and suggest future research directions.

Thanks to the availability and increasing popularity of wearable devices such as GoPro cameras, smart phones, and glasses, we have access to a plethora of videos captured from first person perspective. Surveillance cameras and Unmanned Aerial Vehicles (UAVs) also offer tremendous amounts of video data recorded from top and oblique view points. Egocentric and surveillance vision have been studied extensively but separately in the computer vision community. The relationship between these two domains, however, remains unexplored. In this study, we make the first attempt in this direction by addressing two basic yet challenging questions. First, having a set of egocentric videos and a top-view surveillance video, does the top-view video contain all or some of the egocentric viewers? In other words, have these videos been shot in the same environment at the same time? Second, if so, can we identify the egocentric viewers in the top-view video? These problems can become extremely challenging when videos are not temporally aligned. Each view, egocentric or top, is modeled by a graph and the assignment and time-delays are computed iteratively using the spectral graph matching framework. We evaluate our method in terms of ranking and assigning egocentric viewers to identities present in the top-view video over a dataset of 50 top-view and 188 egocentric videos captured under different conditions. We also evaluate the capability of our proposed approaches in terms of temporal alignment. The experiments and results demonstrate the capability of the proposed approaches in terms of jointly addressing the temporal alignment and assignment tasks.

Recent progress on salient object detection is substantial, benefiting mostly from the explosive development of Convolutional Neural Networks (CNNs). Semantic segmentation and salient object detection algorithms developed lately have been mostly based on Fully Convolutional Neural Networks (FCNs). There is still a large room for improvement over the generic FCN models that do not explicitly deal with the scale-space problem. The Holistically-Nested Edge Detector (HED) provides a skip-layer structure with deep supervision for edge and boundary detection, but the performance gain of HED on saliency detection is not obvious. In this paper, we propose a new salient object detection method by introducing short connections to the skip-layer structures within the HED architecture. Our framework takes full advantage of multi-level and multi-scale features extracted from FCNs, providing more advanced representations at each layer, a property that is critically needed to perform segment detection. Our method produces state-of-the-art results on 5 widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over the existing algorithms. Beyond that, we conduct an exhaustive analysis of the role of training data on performance. We provide a training set for future research and fair comparisons.

A large number of works in egocentric vision have concentrated on action and object recognition. Detection and segmentation of hands in first-person videos, however, has less been explored. For many applications in this domain, it is necessary to accurately segment not only hands of the camera wearer but also the hands of others with whom he is interacting. Here, we take an in-depth look at the hand segmentation problem. In the quest for robust hand segmentation methods, we evaluated the performance of the state of the art semantic segmentation methods, off the shelf and fine-tuned, on existing datasets. We fine-tune RefineNet, a leading semantic segmentation method, for hand segmentation and find that it does much better than the best contenders. Existing hand segmentation datasets are collected in the laboratory settings. To overcome this limitation, we contribute by collecting two new datasets: a) EgoYouTube-Hands including egocentric videos containing hands in the wild, and b) HandOverFace to analyze the performance of our models in presence of similar appearance occlusions. We further explore whether conditional random fields can help refine generated hand segmentations. To demonstrate the benefit of accurate hand maps, we train a CNN for hand-based activity recognition and achieve higher accuracy when a CNN was trained using hand maps produced by the fine-tuned RefineNet. Finally, we annotate a subset of the EgoHands dataset for fine-grained action recognition and show that an accuracy of 58.6% can be achieved by just looking at a single hand pose which is much better than the chance level (12.5%).

Learning to generate natural scenes has always been a challenging task in computer vision. It is even more painstaking when the generation is conditioned on images with drastically different views. This is mainly because understanding, corresponding, and transforming appearance and semantic information across the views is not trivial. In this paper, we attempt to solve the novel problem of cross-view image synthesis, aerial to street-view and vice versa, using conditional generative adversarial networks (cGAN). Two new architectures called Crossview Fork (X-Fork) and Crossview Sequential (X-Seq) are proposed to generate scenes with resolutions of 64 × 64 and 256 × 256 pixels. X-Fork architecture has a single discriminator and a single generator. The generator hallucinates both the image and its semantic segmentation in the target view. X-Seq architecture utilizes two cGANs. The first one generates the target image which is subsequently fed to the second cGAN for generating its corresponding semantic segmentation map. The feedback from the second cGAN helps the first cGAN generate sharper images. Both of our proposed architectures learn to generate natural images as well as their semantic segmentation maps. The proposed methods show that they are able to capture and maintain the true semantics of objects in source and target views better than the traditional image-to-image translation method which considers only the visual appearance of the scene. Extensive qualitative and quantitative evaluations support the effectiveness of our frameworks, compared to two state of the art methods, for natural scene generation across drastically different views.

Research in visual saliency has been focused on two major types of models namely fixation prediction and salient object detection. The relationship between the two, however, has been less explored. In this paper, we propose to employ the former model type to identify and segment salient objects in scenes. We build a novel neural network called Attentive Saliency Network (ASNet) that learns to detect salient objects from fixation maps. The fixation map, derived at the upper network layers, captures a high-level understanding of the scene. Salient object detection is then viewed as fine-grained object-level saliency segmentation and is progressively optimized with the guidance of the fixation map in a top-down manner. ASNet is based on a hierarchy of convolutional LSTMs (convLSTMs) that offers an efficient recurrent mechanism for sequential refinement of the segmentation map. Several loss functions are introduced for boosting the performance of the ASNet. Extensive experimental evaluation shows that our proposed ASNet is capable of generating accurate segmentation maps with the help of the computed fixation map. Our work offers a deeper insight into the mechanisms of attention and narrows the gap between salient object detection and fixation prediction.

In this work, we contribute to video saliency research in two ways. First, we introduce a new benchmark for predicting human eye movements during dynamic scene free-viewing, which is long-time urged in this field. Our dataset, named DHF1K (Dynamic Human Fixation), consists of 1K high-quality, elaborately selected video sequences spanning a large range of scenes, motions, object types and background complexity. Existing video saliency datasets lack variety and generality of common dynamic scenes and fall short in covering challenging situations in unconstrained environments. In contrast, DHF1K makes a significant leap in terms of scalability, diversity and difficulty, and is expected to boost video saliency modeling. Second, we propose a novel video saliency model that augments the CNN-LSTM network architecture with an attention mechanism to enable fast, end-to-end saliency learning. The attention mechanism explicitly encodes static saliency information, thus allowing LSTM to focus on learning more flexible temporal saliency representation across successive frames. Such a design fully leverages existing large-scale static fixation datasets, avoids overfitting, and significantly improves training efficiency and testing performance. We thoroughly examine the performance of our model, with respect to state-of-the-art saliency models, on three large-scale datasets (i.e., DHF1K, Hollywood2, UCF sports). Experimental results over more than 1.2K testing videos containing 400K frames demonstrate that our model outperforms other competitors.

Effective integration of contextual information is crucial for salient object detection. To achieve this, most existing methods based on 'skip' architecture mainly focus on how to integrate hierarchical features of Convolutional Neural Networks (CNNs). They simply apply concatenation or element-wise operation to incorporate high-level semantic cues and low-level detailed information. However, this can degrade the quality of predictions because cluttered and noisy information can also be passed through. To address this problem, we proposes a global Recurrent Localization Network (RLN) which exploits contextual information by the weighted response map in order to localize salient objects more accurately. Particularly, a recurrent module is employed to progressively refine the inner structure of the CNN over multiple time steps. Moreover, to effectively recover object boundaries, we propose a local Boundary Refinement Network (BRN) to adaptively learn the local contextual information for each spatial position. The learned propagation coefficients can be used to optimally capture relations between each pixel and its neighbors. Experiments on five challenging datasets show that our approach performs favorably against all existing methods in terms of the popular evaluation metrics.

The categories and appearance of salient objects vary from image to image, therefore, saliency detection is an image-specific task. Due to lack of large-scale saliency training data, using deep neural networks (DNNs) with pretraining is difficult to precisely capture the image-specific saliency cues. To solve this issue, we formulate a zero-shot learning problem to promote existing saliency detectors. Concretely, a DNN is trained as an embedding function to map pixels and the attributes of the salient/background regions of an image into the same metric space, in which an image-specific classifier is learned to classify the pixels. Since the image-specific task is performed by the classifier, the DNN embedding effectively plays the role of a general feature extractor. Compared with transferring the learning to a new recognition task using limited data, this formulation makes the DNN learn more effectively from small data. Extensive experiments on five data sets show that our method significantly improves accuracy of existing methods and compares favorably against state-of-the-art approaches.

This paper presents a recurrent neural network architecture, guided by the bottom-up attention, for the recognition task. The proposed architecture processes an input image as a sequence of selectively chosen patches. The patches are chosen from the salient regions of the input image. Using human driven saliency maps from gaze, the benefit of such a selection process is first shown. Next, the performance of computational models of bottom-up attention are assessed as alternative to human attention.

The existing binary foreground map (FM) measures address various types of errors in either pixel-wise or structural ways. These measures consider pixel-level match or image-level information independently, while cognitive vision studies have shown that human vision is highly sensitive to both global information and local details in scenes. In this paper, we take a detailed look at current binary FM evaluation measures and propose a novel and effective E-measure (Enhanced-alignment measure). Our measure combines local pixel values with the image-level mean value in one term, jointly capturing image-level statistics and local pixel matching information. We demonstrate the superiority of our measure over the available measures on 4 popular datasets via 5 meta-measures, including ranking models for applications, demoting generic, random Gaussian noise maps, ground-truth switch, as well as human judgments. We find large improvements in almost all the meta-measures. For instance, in terms of application ranking, we observe improvement ranging from 9.08% to 19.65% compared with other popular measures.

In this paper, we attempt to employ convolutional recurrent neural networks for weather temperature estimation using only image data. We study ambient temperature estimation based on deep neural networks in two scenarios a) estimating temperature of a single outdoor image, and b) predicting temperature of the last image in an image sequence. In the first scenario, visual features are extracted by a convolutional neural network trained on a large-scale image dataset. We demonstrate that promising performance can be obtained, and analyze how volume of training data influences performance. In the second scenario, we consider the temporal evolution of visual appearance, and construct a recurrent neural network to predict the temperature of the last image in a given image sequence. We obtain better prediction accuracy compared to the state-of-the-art models. Further, we investigate how performance varies when information is extracted from different scene regions, and when images are captured in different daytime hours. Our approach further reinforces the idea of using only visual information for cost efficient weather prediction in the future.

We propose a novel unsupervised game-theoretic salient object detection algorithm that does not require labeled training data. First, saliency detection problem is formulated as a non-cooperative game, hereinafter referred to as Saliency Game, in which image regions are players who choose to be "background" or "foreground" as their pure strategies. A payoff function is constructed by exploiting multiple cues and combining complementary features. Saliency maps are generated according to each region's strategy in the Nash equilibrium of the proposed Saliency Game. Second, we explore the complementary relationship between color and deep features and propose an Iterative Random Walk algorithm to combine saliency maps produced by the Saliency Game using different features. Iterative random walk allows sharing information across feature spaces, and detecting objects that are otherwise very hard to detect. Extensive experiments over 6 challenging datasets demonstrate the superiority of our proposed unsupervised algorithm compared to several state of the art supervised algorithms.

Co-saliency detection is a newly emerging and rapidly growing research area in the computer vision community. As a novel branch of visual saliency, co-saliency detection refers to the discovery of common and salient foregrounds from two or more relevant images, and it can be widely used in many computer vision tasks. The existing co-saliency detection algorithms mainly consist of three components: extracting effective features to represent the image regions, exploring the informative cues or factors to characterize co-saliency, and designing effective computational frameworks to formulate co-saliency. Although numerous methods have been developed, the literature is still lacking a deep review and evaluation of co-saliency detection techniques. In this article, we aim at providing a comprehensive review of the fundamentals, challenges, and applications of co-saliency detection. Specifically, we provide an overview of some related computer vision works, review the history of co-saliency detection, summarize and categorize the major algorithms in this research area, discuss some open issues in this area, present the potential applications of co-saliency detection, and finally point out some unsolved challenges and promising future works. We expect this review to be beneficial to both fresh and senior researchers in this field and to give insights to researchers in other related areas regarding the utility of co-saliency detection algorithms.

Recent progress on saliency detection is substantial, benefiting mostly from the explosive development of Convolutional Neural Networks (CNNs). Semantic segmentation and saliency detection algorithms developed lately have been mostly based on Fully Convolutional Neural Networks (FCNs). There is still a large room for improvement over the generic FCN models that do not explicitly deal with the scale-space problem. Holisitcally-Nested Edge Detector (HED) provides a skip-layer structure with deep supervision for edge and boundary detection, but the performance gain of HED on saliency detection is not obvious. In this paper, we propose a new saliency method by introducing short connections to the skip-layer structures within the HED architecture. Our framework provides rich multi-scale feature maps at each layer, a property that is critically needed to perform segment detection. Our method produces state-of-the-art results on 5 widely tested salient object detection benchmarks, with advantages in terms of efficiency (0.08 seconds per image), effectiveness, and simplicity over the existing algorithms.

This paper revisits visual saliency prediction by evaluating the recent advancements in this field such as crowd-sourced mouse tracking-based databases and contextual annotations. We pursue a critical and quantitative approach towards some of the new challenges including the quality of mouse tracking versus eye tracking for model training and evaluation. We extend quantitative evaluation of models in order to incorporate contextual information by proposing an evaluation methodology that allows accounting for contextual factors such as text, faces, and object attributes. The proposed contextual evaluation scheme facilitates detailed analysis of models and helps identify their pros and cons. Through several experiments, we find that (1) mouse tracking data has lower inter-participant visual congruency and higher dispersion, compared to the eye tracking data, (2) mouse tracking data does not totally agree with eye tracking in general and in terms of different contextual regions in specific, and (3) mouse tracking data leads to acceptable results in training current existing models, and (4) mouse tracking data is less reliable for model selection and evaluation. The contextual evaluation also reveals that, among the studied models, there is no single model that performs best on all the tested annotations.

Foreground map evaluation is crucial for gauging the progress of object segmentation algorithms, in particular in the field of salient object detection where the purpose is to accurately detect and segment the most salient object in a scene. Several widely-used measures such as Area Under the Curve (AUC), Average Precision (AP) and the recently proposed F W/B (Fbw) have been used to evaluate the similarity between a non-binary saliency map (SM) and a ground-truth (GT) map. These measures are based on pixel-wise errors and often ignore the structural similarities. Behavioral vision studies, however, have shown that the human visual system is highly sensitive to structures in scenes. Here, we propose a novel, efficient, and easy to calculate measure known as structural similarity measure (Structure-measure) to evaluate non-binary foreground maps. Our new measure simultaneously evaluates region-aware and object-aware structural similarity between a SM and a GT map. We demonstrate superiority of our measure over existing ones using 5 meta-measures on 5 benchmark datasets.

To bridge the gap between humans and machines in image understanding and describing, we need further insight into how people describe a perceived scene. In this paper, we study the agreement between bottom-up saliency-based visual attention and object referrals in scene description constructs. We investigate the properties of human-written descriptions and machine-generated ones. We then propose a saliency-boosted image captioning model in order to investigate benefits from low-level cues in language models. We learn that (1) humans mention more salient objects earlier than less salient ones in their descriptions, (2) the better a captioning model performs, the better attention agreement it has with human descriptions, (3) the proposed saliencyboosted model, compared to its baseline form, does not improve significantly on the MS COCO database, indicating explicit bottom-up boosting does not help when the task is well learnt and tuned on a data, (4) a better generalization is, however, observed for the saliency-boosted model on unseen data.

Egocentric, or first-person vision which became popular in recent years with an emerge in wearable technology, is different than exocentric (third-person) vision in some distinguishable ways, one of which being that the camera wearer is generally not visible in the video frames. Recent work has been done on action and object recognition in egocentric videos, as well as work on biometric extraction from first-person videos. Height estimation can be a useful feature for both soft-biometrics and object tracking. Here, we propose a method of estimating the height of an egocentric camera without any calibration or reference points. We used both traditional computer vision approaches and deep learning in order to determine the visual cues that results in best height estimation. Here, we introduce a framework inspired by two stream networks comprising of two Convolutional Neural Networks, one based on spatial information, and one based on information given by optical flow in a frame. Given an egocentric video as an input to the framework, our model yields a height estimate as an output. We also incorporate late fusion to learn a combination of temporal and spatial cues. Comparing our model with other methods we used as baselines, we achieve height estimates for videos with a Mean Average Error of 14.04 cm over a range of 103 cm of data, and classification accuracy for relative height (tall, medium or short) up to 93.75% where chance level is 33%.

Tolerance to image variations (e.g., translation, scale, pose, illumination, background) is an important desired property of any object recognition system, be it human or machine. Moving towards increasingly bigger datasets has been trending in computer vision especially with the emergence of highly popular deep learning models. While being very useful for learning invariance to object inter-and intra-class shape variability, these large-scale wild datasets are not very useful for learning invariance to other parameters urging researchers to resort to other tricks for training models. In this work, we introduce a large-scale synthetic dataset, which is freely and publicly available, and use it to answer several fundamental questions regarding selectivity and invariance properties of convolutional neural networks. Our dataset contains two parts: a) objects shot on a turntable: 15 categories, 8 rotation angles, 11 cameras on a semi-circular arch, 5 lighting conditions, 3 focus levels, variety of backgrounds (23.4 per instance) generating 1320 images per instance (about 22 million images in total), and b) scenes: in which a robotic arm takes pictures of objects on a 1:160 scale scene. We study: 1) invariance and selectivity of different CNN layers, 2) knowledge transfer from one object category to another, 3) systematic or random sampling of images to build a train set, 4) domain adaptation from synthetic to natural scenes, and 5) order of knowledge delivery to CNNs. We also discuss how our analyses can lead the field to develop more efficient deep learning methods.

By predicting where humans look in natural scenes, we can understand how they perceive complex natural scenes and prioritize information for further high-level visual processing. Several models have been proposed for this purpose, yet there is a gap between best existing saliency models and human performance. While many researchers have developed purely computational models for fixation prediction, less attempts have been made to discover cognitive factors that guide gaze. Here, we study the effect of a particular type of scene structural information, known as the vanishing point, and show that human gaze is attracted to the vanishing point regions. We record eye movements of 10 observers over 532 images, out of which 319 have vanishing points. We then construct a combined model of traditional saliency and a vanishing point channel and show that our model outperforms state of the art saliency models using three scores on our dataset.

A large number of saliency models, each based on a different hypothesis, have been proposed over the past 20 years. In practice, while subscribing to one hypothesis or computational principle makes a model that performs well on some types of images, it hinders the general performance of a model on arbitrary images and large-scale data sets. One natural approach to improve overall saliency detection accuracy would then be fusing different types of models. In this paper, inspired by the success of late-fusion strategies in semantic analysis and multi-modal biometrics, we propose to fuse the state-of-the-art saliency models at the score level in a para-boosting learning fashion. First, saliency maps generated by several models are used as confidence scores. Then, these scores are fed into our para-boosting learner (i.e., support vector machine, adaptive boosting, or probability density estimator) to generate the final saliency map. In order to explore the strength of para-boosting learners, traditional transformation-based fusion strategies, such as Sum, Min, and Max, are also explored and compared in this paper. To further reduce the computation cost of fusing too many models, only a few of them are considered in the next step. Experimental results show that score-level fusion outperforms each individual model and can further reduce the performance gap between the current models and the human inter-observer model.

Predicting where people look in natural scenes has attracted a lot of interest in computer vision and computational neuroscience over the past two decades. Two seemingly contrasting categories of cues have been proposed to influence where people look: 1) low-level image saliency and 2) high-level semantic information. Our first contribution is to take a detailed look at these cues to confirm the hypothesis proposed by Henderson and Nuthmann and Henderson that observers tend to look at the center of objects. We analyzed fixation data for scene free-viewing over 17 observers on 60 object-annotated images with various types of objects. Images contained different types of scenes, such as natural scenes, line drawings, and 3-D rendered scenes. Our second contribution is to propose a simple combined model of low-level saliency and object center bias that outperforms each individual component significantly over our data, as well as on the Object and Semantic Images and Eye-tracking data set by Xu et al. The results reconcile saliency with object center-bias hypotheses and highlight that both types of cues are important in guiding fixations. Our work opens new directions to understand strategies that humans use in observing scenes and objects, and demonstrates the construction of combined models of low-level saliency and high-level object-based information.

Advances in image quality assessment have shown the potential added value of including visual attention aspects in its objective assessment. Numerous models of visual saliency are implemented and integrated in different image quality metrics (IQMs), but the gain in reliability of the resulting IQMs varies to a large extent. The causes and the trends of this variation would be highly beneficial for further improvement of IQMs, but are not fully understood. In this paper, an exhaustive statistical evaluation is conducted to justify the added value of computational saliency in objective image quality assessment, using 20 state-of-the-art saliency models and 12 best-known IQMs. Quantitative results show that the difference in predicting human fixations between saliency models is sufficient to yield a significant difference in performance gain when adding these saliency models to IQMs. However, surprisingly, the extent to which an IQM can profit from adding a saliency model does not appear to have direct relevance to how well this saliency model can predict human fixations. Our statistical analysis provides useful guidance for applying saliency models in IQMs, in terms of the effect of saliency model dependence, IQM dependence, and image distortion dependence. The testbed and software are made publicly available to the research community.

Salient object detection remains one of the most important and active research topics in computer vision, with wide-ranging applications to object recognition, scene understanding, image retrieval, context aware image editing, image compression, etc. Most existing methods directly determine salient objects by exploring various salient object features. Here, we propose a novel graph based ranking method to detect and segment the most salient object in a scene according to its relationship to image border (background) regions, i.e., the background feature. Firstly, we use regions/super-pixels as graph nodes, which are fully connected to enable both long range and short range relations to be modeled. The relationship of each region to the image border (background) is evaluated in two stages: (i) ranking with hard background queries, and (ii) ranking with soft foreground queries. We experimentally show how this two-stage ranking based salient object detection method is complementary to traditional methods, and that integrated results outperform both. Our method allows the exploitation of intrinsic image structure to achieve high quality salient object determination using a quadratic optimization framework, with a closed form solution which can be easily computed. Extensive method evaluation and comparison using three challenging saliency datasets demonstrate that our method consistently outperforms 10 state-of-the-art models by a big margin.

Salient object detection or salient region detection models, diverging from fixation prediction models, have traditionally been dealing with locating and segmenting the most salient object or region in a scene. While the notion of most salient object is sensible when multiple objects exist in a scene, current datasets for evaluation of saliency detection approaches often have scenes with only one single object. We introduce three main contributions in this paper. First, we take an in-depth look at the problem of salient object detection by studying the relationship between where people look in scenes and what they choose as the most salient object when they are explicitly asked. Based on the agreement between fixations and saliency judgments, we then suggest that the most salient object is the one that attracts the highest fraction of fixations. Second, we provide two new less biased benchmark data sets containing scenes with multiple objects that challenge existing saliency models. Indeed, we observed a severe drop in performance of eight state-of-the-art models on our data sets (40%-70%). Third, we propose a very simple yet powerful model based on superpixels to be used as a baseline for model evaluation and comparison. While on par with the best models on MSRA-5 K data set, our model wins over other models on our data highlighting a serious drawback of existing models, which is convoluting the processes of locating the most salient object and its segmentation. We also provide a review and statistical analysis of some labeled scene data sets that can be used for evaluating salient object detection models. We believe that our work can greatly help remedy the over-fitting of models to existing biased data sets and opens new venues for future research in this fast-evolving field.

We extensively compare, qualitatively and quantitatively, 41 state-of-the-art models (29 salient object detection, 10 fixation prediction, 1 objectness, and 1 baseline) over seven challenging data sets for the purpose of benchmarking salient object detection and segmentation methods. From the results obtained so far, our evaluation shows a consistent rapid progress over the last few years in terms of both accuracy and running time. The top contenders in this benchmark significantly outperform the models identified as the best in the previous benchmark conducted three years ago. We find that the models designed specifically for salient object detection generally work better than models in closely related areas, which in turn provides a precise definition and suggests an appropriate treatment of this problem that distinguishes it from other problems. In particular, we analyze the influences of center bias and scene complexity in model performance, which, along with the hard cases for the state-of-the-art models, provide useful hints toward constructing more challenging large-scale data sets and better saliency models. Finally, we propose probable solutions for tackling several open problems, such as evaluation scores and data set bias, which also suggest future research directions in the rapidly growing field of salient object detection.

Several decades of research in computer and primate vision have resulted in many models (some specialized for one problem, others more general) and invaluable experimental data. Here, to help focus research efforts onto the hardest unsolved problems, and bridge computer and human vision, we define a battery of 5 tests that measure the gap between human and machine performances in several dimensions (generalization across scene categories, generalization from images to edge maps and line drawings, invariance to rotation and scaling, local/global information with jumbled images, and object recognition performance). We measure model accuracy and the correlation between model and human error patterns. Experimenting over 7 datasets, where human data is available, and gauging 14 well-established models, we find that none fully resembles humans in all aspects, and we learn from each test which models and features are more promising in approaching humans in the tested dimension. Across all tests, we find that models based on local edge histograms consistently resemble humans more, while several scene statistics or "gist" models do perform well with both scenes and objects. While computer vision has long been inspired by human vision, we believe systematic efforts, such as this, will help better identify shortcomings of models and find new paths forward.

Advances in image quality assessment have shown the potential added value of including visual attention aspects in objective quality metrics. Numerous models of visual saliency are implemented and integrated in different quality metrics; however, their ability of improving a metric's performance in predicting perceived image quality is not fully investigated. In this paper, we conduct an exhaustive comparison of 20 state-of-the-art saliency models in the context of image quality assessment. Experimental results show that adding computational saliency is beneficial to quality prediction in general terms. However, the amount of performance gain that can be obtained by adding saliency in quality metrics highly depends on the saliency model and on the metric.

Several visual attention models have been proposed for describing eye movements over simple stimuli and tasks such as free viewing or visual search. Yet, to date, there exists no computational framework that can reliably mimic human gaze behavior in more complex environments and tasks such as urban driving. In addition, benchmark datasets, scoring techniques, and top-down model architectures are not yet well understood. In this paper, we describe new task-dependent approaches for modeling top-down overt visual attention based on graphical models for probabilistic inference and reasoning. We describe a dynamic Bayesian network that infers probability distributions over attended objects and spatial locations directly from observed data. Probabilistic inference in our model is performed over object-related functions that are fed from manual annotations of objects in video scenes or by state-of-the-art object detection/recognition algorithms. Evaluating over approximately 3 h (approximately 315 000 eye fixations and 12 000 saccades) of observers playing three video games (time-scheduling, driving, and flight combat), we show that our approach is significantly more predictive of eye fixations compared to: 1) simpler classifier-based models also developed here that map a signature of a scene (multimodal information from gist, bottom-up saliency, physical actions, and events) to eye positions; 2) 14 state-of-the-art bottom-up saliency models; and 3) brute-force algorithms such as mean eye position. Our results show that the proposed model is more effective in employing and reasoning over spatio-temporal visual data compared with the state-of-the-art.

Significant recent progress has been made in developing high-quality saliency models. However, less effort has been undertaken on fair assessment of these models, over large standardized datasets and correctly addressing confounding factors. In this study, we pursue a critical and quantitative look at challenges (e.g., center-bias, map smoothing) in saliency modeling and the way they affect model accuracy. We quantitatively compare 32 state-of-the-art models (using the shuffled AUC score to discount center-bias) on 4 benchmark eye movement datasets, for prediction of human fixation locations and scan path sequence. We also account for the role of map smoothing. We find that, although model rankings vary, some (e.g., AWS, LG, AIM, and HouNIPS) consistently outperform other models over all datasets. Some models work well for prediction of both fixation locations and scan path sequence (e.g., Judd, GBVS). Our results show low prediction accuracy for models over emotional stimuli from the NUSEF dataset. Our last benchmark, for the first time, gauges the ability of models to decode the stimulus category from statistics of fixations, saccades, and model saliency values at fixated locations. In this test, ITTI and AIM models win over other models. Our benchmark provides a comprehensive high-level picture of the strengths and weaknesses of many popular models, and suggests future research directions in saliency modeling.

Many real-world problems have complicated objective functions. To optimize such functions, humans utilize sophisticated sequential decision-making strategies. Many optimization algorithms have also been developed for this same purpose, but how do they compare to humans in terms of both performance and behavior? We try to unravel the general underlying algorithm people may be using while searching for the maximum of an invisible 1D function. Subjects click on a blank screen and are shown the ordinate of the function at each clicked abscissa location. Their task is to find the function’s maximum in as few clicks as possible. Subjects win if they get close enough to the maximum location. Analysis over 23 non-maths undergraduates, optimizing 25 functions from different families, shows that humans outperform 24 well-known optimization algorithms. Bayesian Optimization based on Gaussian Processes, which exploit all the x values tried and all the f(x) values obtained so far to pick the next x, predicts human performance and searched locations better. In 6 follow-up controlled experiments over 76 subjects, covering interpolation, extrapolation, and optimization tasks, we further confirm that Gaussian Processes provide a general and unified theoretical account to explain passive and active function learning and search in humans.

Modeling visual attention--particularly stimulus-driven, saliency-based attention--has been a very active research area over the past 25 years. Many different models of attention are now available which, aside from lending theoretical contributions to other fields, have demonstrated successful applications in computer vision, mobile robotics, and cognitive systems. Here we review, from a computational perspective, the basic concepts of attention implemented in these models. We present a taxonomy of nearly 65 models, which provides a critical comparison of approaches, their capabilities, and shortcomings. In particular, 13 criteria derived from behavioral and computational studies are formulated for qualitative comparison of attention models. Furthermore, we address several challenging issues with models, including biological plausibility of the computations, correlation with eye movement datasets, bottom-up and top-down dissociation, and constructing meaningful performance measures. Finally, we highlight current research trends in attention modeling and provide insights for future.

Visual attention is a process that enables biological and machine vision systems to select the most relevant regions from a scene. Relevance is determined by two components: 1) top-down factors driven by task and 2) bottom-up factors that highlight image regions that are different from their surroundings. The latter are often referred to as "visual saliency." Modeling bottom-up visual saliency has been the subject of numerous research efforts during the past 20 years, with many successful applications in computer vision and robotics. Available models have been tested with different datasets (e.g., synthetic psychological search arrays, natural images or videos) using different evaluation scores (e.g., search slopes, comparison to human eye tracking) and parameter settings. This has made direct comparison of models difficult. Here, we perform an exhaustive comparison of 35 state-of-the-art saliency models over 54 challenging synthetic patterns, three natural image datasets, and two video datasets, using three evaluation scores. We find that although model rankings vary, some models consistently perform better. Analysis of datasets reveals that existing datasets are highly center-biased, which influences some of the evaluation scores. Computational complexity analysis shows that some models are very fast, yet yield competitive eye movement prediction accuracy. Different models often have common easy/difficult stimuli. Furthermore, several concerns in visual saliency modeling, eye movement datasets, and evaluation scores are discussed and insights for future work are provided. Our study allows one to assess the state-of-the-art, helps to organizing this rapidly growing field, and sets a unified comparison framework for gauging future efforts, similar to the PASCAL VOC challenge in the object recognition and detection domains.

We introduce a new task-independent framework to model top-down overt visual attention based on graph-ical models for probabilistic inference and reasoning. We describe a Dynamic Bayesian Network (DBN) that infers probability distributions over attended objects and spatial locations directly from observed data. Probabilistic inference in our model is performed over object-related functions which are fed from manual annotations of objects in video scenes or by state-of-the-art object detection models. Evaluating over ∼3 hours (appx. 315,000 eye fixations and 12,600 saccades) of observers playing 3 video games (time-scheduling, driving, and flight combat), we show that our approach is significantly more predictive of eye fixations compared to: 1) simpler classifier-based models also developed here that map a signature of a scene (multi-modal information from gist, bottom-up saliency, physical actions, and events) to eye positions, 2) 14 state-of-the-art bottom-up saliency models, and 3) brute-force algorithms such as mean eye position. Our results show that the proposed model is more effective in employing and reasoning over spatio-temporal visual data.

Despite significant recent progress, the best available visual saliency models still lag behind human performance in predicting eye fixations in free-viewing of natural scenes. Majority of models are based on low-level visual features and the importance of top-down factors has not yet been fully explored or modeled. Here, we combine low-level features such as orientation, color, intensity, saliency maps of previous best bottom-up models with top-down cognitive visual features (e.g., faces, humans, cars, etc.) and learn a direct mapping from those features to eye fixations using Regression, SVM, and AdaBoost classifiers. By extensive experimenting over three benchmark eye-tracking datasets using three popular evaluation scores, we show that our boosting model outperforms 27 state-of-the-art models and is so far the closest model to the accuracy of human model for fixation prediction. Furthermore, our model successfully detects the most salient object in a scene without sophisticated image processings such as region segmentation.

We introduce a saliency model based on two key ideas. The first one is considering local and global image patch rarities as two complementary processes. The second one is based on our observation that for different images, one of the RGB and Lab color spaces outperforms the other in saliency detection. We propose a framework that measures patch rarities in each color space and combines them in a final map. For each color channel, first, the input image is partitioned into non-overlapping patches and then each patch is represented by a vector of coefficients that linearly reconstruct it from a learned dictionary of patches from natural scenes. Next, two measures of saliency (Local and Global) are calculated and fused to indicate saliency of each patch. Local saliency is distinctiveness of a patch from its surrounding patches. Global saliency is the inverse of a patch's probability of happening over the entire image. The final saliency map is built by normalizing and fusing local and global saliency maps of all channels from both color systems. Extensive evaluation over four benchmark eye-tracking datasets shows the significant advantage of our approach over 10 state-of-the-art saliency models.

Despite a considerable amount of previous work on bottom-up saliency modeling for predicting human fixations over static and dynamic stimuli, few studies have thus far attempted to model top-down and task-driven influences of visual attention. Here, taking advantage of the sequential nature of real-world tasks, we propose a unified Bayesian approach for modeling task-driven visual attention. Several sources of information, including global context of a scene, previous attended locations, and previous motor actions, are integrated over time to predict the next attended location. Recording eye movements while subjects engage in 5 contemporary 2D and 3D video games, as modest counterparts of everyday tasks, we show that our approach is able to predict human attention and gaze better than the state-of-the-art, with a large margin (about 15% increase in prediction accuracy). The advantage of our approach is that it is automatic and applicable to arbitrary visual tasks.

A large number of studies have been reported on top-down influences of visual attention. However, less progress have been made in understanding and modeling its mechanisms in real-world tasks. In this paper, we propose an approach for learning spatial attention taking into account influences of physical actions on top-down attention. For this purpose, we focus on interactive visual environments (video games) which are modest real-world simulations, where a player has to attend to certain aspects of visual stimuli and perform actions to achieve a goal. The basic idea is to learn a mapping from current mental state of the game player, represented by past actions and observations, to its gaze fixation. A data-driven approach is followed where we train a model from the data of some players and test it over a new subject. In particular, two contributions this paper makes are: 1) employing multi-modal information including mean eye position, gist of a scene, physical actions, bottom-up saliency, and tagged events for state representation and 2) analysis of different methods of combining bottom-up and top-down influences. Comparing with other top-down task-driven and bottom-up spatio-temporal models, our approach shows higher NSS scores in predicting eye positions.

Modeling how visual saliency guides the deployment of attention over visual scenes has attracted much interest recently — among both computer vision and experimental/computational researchers — since visual attention is a key function of both machine and biological vision systems. Research efforts in computer vision have mostly been focused on modeling bottom-up saliency. Strong influences o n attention and eye movements, however, come from instantaneous task demands. Here, we propose models of top-down visual guidance considering task influences. The n ew models estimate the state of a human subject performing a task (here, playing video games), and map that state to an eye position. Factors influencing state come from scene gi st, physical actions, events, and bottom-up saliency. Proposed models fall into two categories. In the first category, we use classical discriminative classifiers, including Reg ression, kNN and SVM. In the second category, we use Bayesian Networks to combine all the multi-modal factors in a unified framework. Our approaches significantly outperfor m 15 competing bottom-up and top-down attention models in predicting future eye fixat ions on 18,000 and 75,00 video frames and eye movement samples from a driving and a flig ht combat video game, respectively. We further test and validate our approaches on 1.4M video frames and 11M fixations samples and in all cases obtain higher prediction s cores that reference models.

This work proposes an approach for scene classification by extracting and matching visual features only at the focuses of visual attention instead of the entire scene. Analysis over a database of natural scenes demonstrates that regions proposed by the saliency-based model of visual attention are robust to image transformations. Using a nearest neighbor classifier and a distance measure defined over the salient regions, we obtained 97.35% and 78.28% classification rates with SIFT and C2 features from the HMAX model at 5 salient regions covering at most 31% of the image. Classification with features extracted from the entire image results in 99.3% and 82.32% using SIFT and C2 features, respectively. Comparing attentional and adhoc approaches shows that classification rate of the first approach is 0.95 of the second. Overall, our results prove that efficient scene classification, in terms of reducing the complexity of feature extraction is possible without a significant drop in performance.

A major problem with previous object tracking approaches is adapting object representations depending on scene context to account for changes in illumination, viewpoint changes, etc. To adapt our previous approach to deal with background changes, here we first derive some clusters from a training sequence and the corresponding object representations for those clusters. Next, for each frame of a separate test sequence, its nearest background cluster is determined and then the corresponding descriptor of that cluster is used for object representation in this frame. Experiments show that the proposed approach tracks objects and persons in natural scenes more effectively.

This paper introduces a new method for learning top-down and task-driven visual attention control along with physical actions in interactive environments. Our method is based on the Reinforcement Learning of Visual Classes(RLVC) algorithm and adapts it for learning spatial visual selection in order to reduce computational complexity. Proposed algorithm also addresses aliasings due to not knowing previous actions and perceptions. Continuing learning shows our method is robust to perturbations in perceptual information. Our method also allows object recognition when class labels are used instead of physical actions. We have tried to gain maximum generalization while performing local processing. Experiments over visual navigation and object recognition tasks show that our method is more efficient in terms of computational complexity and is biologically more plausible.

Similar to humans and primates, artificial creatures like robots are limited in terms of allocation of their resources to huge sensory and perceptual information. Serial processing mechanisms used in the design of such creatures demands engineering attentional control mechanisms. In this paper, we present a new algorithm for learning top-down sequential visual attention control for agents acting in interactive environments. Our method is based on the key idea, that attention can be learned best in concert with visual representations through automatic construction and discretization of the visual state space. The tree representing the top-down attention is incrementally refined whenever aliasing occurs by selecting the most appropriate saccadic direction. The proposed approach is evaluated on action-based object recognition and urban navigation tasks, where obtained results support applicability and usefulness of developed saccade movement method for robotics.

In this paper, we propose a fast algorithm for gesture recognition based on the saliency maps of visual attention. A tuned saliency-based model of visual attention is used to find potential hand regions in video frames. To obtain the overall movement of the hand, saliency maps of the differences of consecutive video frames are overlaid. An improved characteristic loci feature extraction method is introduced and used to code obtained hand movement. Finally, the extracted feature vector is used for training SVMs to classify the gestures. The proposed method along a hand-eye coordination model is used to play a robotic marionette and an approval/rejection phase is used to interactively correct the robotic marionette's behavior.