Current Deep Neural Networks are vulnerable to adversarial examples, which alter their predictions by adding carefully crafted noise. Since human eyes are robust to such inputs, it is possible that the vulnerability stems from the standard way of processing inputs in one shot by processing every pixel with the same importance. In contrast, neuroscience suggests that the human vision system can differentiate salient features by (1) switching between multiple fixation points (saccades) and (2) processing the surrounding with a non-uniform external resolution (foveation). In this work, we advocate that the integration of such active vision mechanisms into current deep learning systems can offer robustness benefits. Specifically, we empirically demonstrate the inherent robustness of two active vision methods - GFNet and FALcon - under a black box threat model. By learning and inferencing based on downsampled glimpses obtained from multiple distinct fixation points within an input, we show that these active methods achieve (2-3) times greater robustness compared to a standard passive convolutional network under state-of-the-art adversarial attacks. More importantly, we provide illustrative and interpretable visualization analysis that demonstrates how performing inference from distinct fixation points makes active vision methods less vulnerable to malicious inputs.
Deep neural network (DNN) based machine perception frameworks process the entire input in a one-shot manner to provide answers to both "what object is being observed" and "where it is located". In contrast, the "two-stream hypothesis" from neuroscience explains the neural processing in the human visual cortex as an active vision system that utilizes two separate regions of the brain to answer the what and the where questions. In this work, we propose a machine learning framework inspired by the "two-stream hypothesis" and explore the potential benefits that it offers. Specifically, the proposed framework models the following mechanisms: 1) ventral (what) stream focusing on the input regions perceived by the fovea part of an eye (foveation), 2) dorsal (where) stream providing visual guidance, and 3) iterative processing of the two streams to calibrate visual focus and process the sequence of focused image patches. The training of the proposed framework is accomplished by label-based DNN training for the ventral stream model and reinforcement learning for the dorsal stream model. We show that the two-stream foveation-based learning is applicable to the challenging task of weakly-supervised object localization (WSOL), where the training data is limited to the object class or its attributes. The framework is capable of both predicting the properties of an object and successfully localizing it by predicting its bounding box. We also show that, due to the independent nature of the two streams, the dorsal model can be applied on its own to unseen images to localize objects from different datasets.
Foveated vision, a trait shared by many animals, including humans, has not been fully utilized in machine learning applications, despite its significant contributions to biological visual function. This study investigates whether retinotopic mapping, a critical component of foveated vision, can enhance image categorization and localization performance when integrated into deep convolutional neural networks (CNNs). Retinotopic mapping was integrated into the inputs of standard off-the-shelf convolutional neural networks (CNNs), which were then retrained on the ImageNet task. As expected, the logarithmic-polar mapping improved the network's ability to handle arbitrary image zooms and rotations, particularly for isolated objects. Surprisingly, the retinotopically mapped network achieved comparable performance in classification. Furthermore, the network demonstrated improved classification localization when the foveated center of the transform was shifted. This replicates a crucial ability of the human visual system that is absent in typical convolutional neural networks (CNNs). These findings suggest that retinotopic mapping may be fundamental to significant preattentive visual processes.
In this paper, we tackle the challenge of actively attending to visual scenes using a foveated sensor. We introduce an end-to-end differentiable foveated active vision architecture that leverages a graph convolutional network to process foveated images, and a simple yet effective formulation for foveated image sampling. Our model learns to iteratively attend to regions of the image relevant for classification. We conduct detailed experiments on a variety of image datasets, comparing the performance of our method with previous approaches to foveated vision while measuring how the impact of different choices, such as the degree of foveation, and the number of fixations the network performs, affect object recognition performance. We find that our model outperforms a state-of-the-art CNN and foveated vision architectures of comparable parameters and a given pixel or computation budget
Fourier single-pixel imaging (FSI) is a data-efficient single-pixel imaging (SPI). However, there is still a serious challenge to obtain higher imaging quality using fewer measurements, which limits the development of real-time SPI. In this work, a uniform-sampling foveated FSI (UFFSI) is proposed with three features, uniform sampling, effective sampling and flexible fovea, to achieve under-sampling high-efficiency and high-quality SPI, even in a large-scale scene. First, by flexibly using the three proposed foveated pattern structures, data redundancy is reduced significantly to only require high resolution (HR) on regions of interest (ROIs), which radically reduces the need of total data number. Next, by the non-uniform weight distribution processing, non-uniform spatial sampling is transformed into uniform sampling, then the fast Fourier transform is used accurately and directly to obtain under-sampling high imaging quality with further reduced measurements. At a sampling ratio of 0.0084 referring to HR FSI with 1024*768 pixels, experimentally, by UFFSI with 255*341 cells of 89% reduction in data redundancy, the ROI has a significantly better imaging quality to meet imaging needs. We hope this work can provide a breakthrough for future real-time SPI.
Deep neural networks (DNNs) have been shown to be vulnerable to adversarial attacks -- subtle, perceptually indistinguishable perturbations of inputs that change the response of the model. In the context of vision, we hypothesize that an important contributor to the robustness of human visual perception is constant exposure to low-fidelity visual stimuli in our peripheral vision. To investigate this hypothesis, we develop \RBlur, an image transform that simulates the loss in fidelity of peripheral vision by blurring the image and reducing its color saturation based on the distance from a given fixation point. We show that compared to DNNs trained on the original images, DNNs trained on images transformed by \RBlur are substantially more robust to adversarial attacks, as well as other, non-adversarial, corruptions, achieving up to 25\% higher accuracy on perturbed data.
Understanding the mechanisms underlying human attention is a fundamental challenge for both vision science and artificial intelligence. While numerous computational models of free-viewing have been proposed, less is known about the mechanisms underlying task-driven image exploration. To address this gap, we present CapMIT1003, a database of captions and click-contingent image explorations collected during captioning tasks. CapMIT1003 is based on the same stimuli from the well-known MIT1003 benchmark, for which eye-tracking data under free-viewing conditions is available, which offers a promising opportunity to concurrently study human attention under both tasks. We make this dataset publicly available to facilitate future research in this field. In addition, we introduce NevaClip, a novel zero-shot method for predicting visual scanpaths that combines contrastive language-image pretrained (CLIP) models with biologically-inspired neural visual attention (NeVA) algorithms. NevaClip simulates human scanpaths by aligning the representation of the foveated visual stimulus and the representation of the associated caption, employing gradient-driven visual exploration to generate scanpaths. Our experimental results demonstrate that NevaClip outperforms existing unsupervised computational models of human visual attention in terms of scanpath plausibility, for both captioning and free-viewing tasks. Furthermore, we show that conditioning NevaClip with incorrect or misleading captions leads to random behavior, highlighting the significant impact of caption guidance in the decision-making process. These findings contribute to a better understanding of mechanisms that guide human attention and pave the way for more sophisticated computational approaches to scanpath prediction that can integrate direct top-down guidance of downstream tasks.
The success of deep learning is frequently described as the ability to train all parameters of a network on a specific application in an end-to-end fashion. Yet, several design choices on the camera level, including the pixel layout of the sensor, are considered as pre-defined and fixed, and high resolution, regular pixel layouts are considered to be the most generic ones in computer vision and graphics, treating all regions of an image as equally important. While several works have considered non-uniform, \eg, hexagonal or foveated, pixel layouts in hardware and image processing, the layout has not been integrated into the end-to-end learning paradigm so far. In this work, we present the first truly end-to-end trained imaging pipeline that optimizes the size and distribution of pixels on the imaging sensor jointly with the parameters of a given neural network on a specific task. We derive an analytic, differentiable approach for the sensor layout parameterization that allows for task-specific, local varying pixel resolutions. We present two pixel layout parameterization functions: rectangular and curvilinear grid shapes that retain a regular topology. We provide a drop-in module that approximates sensor simulation given existing high-resolution images to directly connect our method with existing deep learning models. We show that network predictions benefit from learnable pixel layouts for two different downstream tasks, classification and semantic segmentation.
The human visual system processes images with varied degrees of resolution, with the fovea, a small portion of the retina, capturing the highest acuity region, which gradually declines toward the field of view's periphery. However, the majority of existing object localization methods rely on images acquired by image sensors with space-invariant resolution, ignoring biological attention mechanisms. As a region of interest pooling, this study employs a fixation prediction model that emulates human objective-guided attention of searching for a given class in an image. The foveated pictures at each fixation point are then classified to determine whether the target is present or absent in the scene. Throughout this two-stage pipeline method, we investigate the varying results obtained by utilizing high-level or panoptic features and provide a ground-truth label function for fixation sequences that is smoother, considering in a better way the spatial structure of the problem. Finally, we present a novel dual task model capable of performing fixation prediction and detection simultaneously, allowing knowledge transfer between the two tasks. We conclude that, due to the complementary nature of both tasks, the training process benefited from the sharing of knowledge, resulting in an improvement in performance when compared to the previous approach's baseline scores.
Most models of visual attention are aimed at predicting either top-down or bottom-up control, as studied using different visual search and free-viewing tasks. We propose Human Attention Transformer (HAT), a single model predicting both forms of attention control. HAT is the new state-of-the-art (SOTA) in predicting the scanpath of fixations made during target-present and target-absent search, and matches or exceeds SOTA in the prediction of taskless free-viewing fixation scanpaths. HAT achieves this new SOTA by using a novel transformer-based architecture and a simplified foveated retina that collectively create a spatio-temporal awareness akin to the dynamic visual working memory of humans. Unlike previous methods that rely on a coarse grid of fixation cells and experience information loss due to fixation discretization, HAT features a dense-prediction architecture and outputs a dense heatmap for each fixation, thus avoiding discretizing fixations. HAT sets a new standard in computational attention, which emphasizes both effectiveness and generality. HAT's demonstrated scope and applicability will likely inspire the development of new attention models that can better predict human behavior in various attention-demanding scenarios.