This paper presents Segment This Thing (STT), a new efficient image segmentation model designed to produce a single segment given a single point prompt. Instead of following prior work and increasing efficiency by decreasing model size, we gain efficiency by foveating input images. Given an image and a point prompt, we extract a crop centered on the prompt and apply a novel variable-resolution patch tokenization in which patches are downsampled at a rate that increases with increased distance from the prompt. This approach yields far fewer image tokens than uniform patch tokenization. As a result we can drastically reduce the computational cost of segmentation without reducing model size. Furthermore, the foveation focuses the model on the region of interest, a potentially useful inductive bias. We show that our Segment This Thing model is more efficient than prior work while remaining competitive on segmentation benchmarks. It can easily run at interactive frame rates on consumer hardware and is thus a promising tool for augmented reality or robotics applications.
The true vision for real-time virtual and augmented reality is reproducing our visual reality in its entirety on immersive displays. To this end, foveated rendering leverages the limitations of spatial acuity in human peripheral vision to allocate computational resources to the fovea while reducing quality in the periphery. Such methods are often derived from studies on the spatial resolution of the human visual system and its ability to perceive blur in the periphery, enabling the potential for high spatial quality in real-time. However, the effects of blur on other visual cues that depend on luminance contrast, such as depth, remain largely unexplored. It is critical to understand this interplay, as accurate depth representation is a fundamental aspect of visual realism. In this paper, we present the first evaluation exploring the effects of foveated rendering on stereoscopic depth perception. We design a psychovisual experiment to quantitatively study the effects of peripheral blur on depth perception. Our analysis demonstrates that stereoscopic acuity remains unaffected (or even improves) by high levels of peripheral blur. Based on our studies, we derive a simple perceptual model that determines the amount of foveation that does not affect stereoacuity. Furthermore, we analyze the model in the context of common foveation practices reported in literature. The findings indicate that foveated rendering does not impact stereoscopic depth perception, and stereoacuity remains unaffected with up to 2x stronger foveation than commonly used. Finally, we conduct a validation experiment and show that our findings hold for complex natural stimuli.
We introduce Foveation-based Explanations (FovEx), a novel human-inspired visual explainability (XAI) method for Deep Neural Networks. Our method achieves state-of-the-art performance on both transformer (on 4 out of 5 metrics) and convolutional models (on 3 out of 5 metrics), demonstrating its versatility. Furthermore, we show the alignment between the explanation map produced by FovEx and human gaze patterns (+14\% in NSS compared to RISE, +203\% in NSS compared to gradCAM), enhancing our confidence in FovEx's ability to close the interpretation gap between humans and machines.
The objects we perceive guide our eye movements when observing real-world dynamic scenes. Yet, gaze shifts and selective attention are critical for perceiving details and refining object boundaries. Object segmentation and gaze behavior are, however, typically treated as two independent processes. Here, we present a computational model that simulates these processes in an interconnected manner and allows for hypothesis-driven investigations of distinct attentional mechanisms. Drawing on an information processing pattern from robotics, we use a Bayesian filter to recursively segment the scene, which also provides an uncertainty estimate for the object boundaries that we use to guide active scene exploration. We demonstrate that this model closely resembles observers' free viewing behavior on a dataset of dynamic real-world scenes, measured by scanpath statistics, including foveation duration and saccade amplitude distributions used for parameter fitting and higher-level statistics not used for fitting. These include how object detections, inspections, and returns are balanced and a delay of returning saccades without an explicit implementation of such temporal inhibition of return. Extensive simulations and ablation studies show that uncertainty promotes balanced exploration and that semantic object cues are crucial to forming the perceptual units used in object-based attention. Moreover, we show how our model's modular design allows for extensions, such as incorporating saccadic momentum or pre-saccadic attention, to further align its output with human scanpaths.
Multimodal medical image fusion (MMIF) is highly significant in such fields as disease diagnosis and treatment. The traditional MMIF methods are difficult to provide satisfactory fusion accuracy and robustness due to the influence of such possible human-crafted components as image transform and fusion strategies. Existing deep learning based fusion methods are generally difficult to ensure image fusion effect due to the adoption of a human-designed network structure and a relatively simple loss function and the ignorance of human visual characteristics during weight learning. To address these issues, we have presented the foveated differentiable architecture search (F-DARTS) based unsupervised MMIF method. In this method, the foveation operator is introduced into the weight learning process to fully explore human visual characteristics for the effective image fusion. Meanwhile, a distinctive unsupervised loss function is designed for network training by integrating mutual information, sum of the correlations of differences, structural similarity and edge preservation value. Based on the presented foveation operator and loss function, an end-to-end encoder-decoder network architecture will be searched using the F-DARTS to produce the fused image. Experimental results on three multimodal medical image datasets demonstrate that the F-DARTS performs better than several traditional and deep learning based fusion methods by providing visually superior fused results and better objective evaluation metrics.
If neuroscientists were asked which brain area is responsible for object recognition in primates, most would probably answer infero-temporal (IT) cortex. While IT is likely responsible for fine discriminations, and it is accordingly dominated by foveal visual inputs, there is more to object recognition than fine discrimination. Importantly, foveation of an object of interest usually requires recognizing, with reasonable confidence, its presence in the periphery. Arguably, IT plays a secondary role in such peripheral recognition, and other visual areas might instead be more critical. To investigate how signals carried by early visual processing areas (such as LGN and V1) could be used for object recognition in the periphery, we focused here on the task of distinguishing faces from non-faces. We tested how sensitive various models were to nuisance parameters, such as changes in scale and orientation of the image, and the type of image background. We found that a model of V1 simple or complex cells could provide quite reliable information, resulting in performance better than 80% in realistic scenarios. An LGN model performed considerably worse. Because peripheral recognition is both crucial to enable fine recognition (by bringing an object of interest on the fovea), and probably sufficient to account for a considerable fraction of our daily recognition-guided behavior, we think that the current focus on area IT and foveal processing is too narrow. We propose that rather than a hierarchical system with IT-like properties as its primary aim, object recognition should be seen as a parallel process, with high-accuracy foveal modules operating in parallel with lower-accuracy and faster modules that can operate across the visual field.
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.
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
Humans actively observe the visual surroundings by focusing on salient objects and ignoring trivial details. However, computer vision models based on convolutional neural networks (CNN) often analyze visual input all at once through a single feed-forward pass. In this study, we designed a dual-stream vision model inspired by the human brain. This model features retina-like input layers and includes two streams: one determining the next point of focus (the fixation), while the other interprets the visuals surrounding the fixation. Trained on image recognition, this model examines an image through a sequence of fixations, each time focusing on different parts, thereby progressively building a representation of the image. We evaluated this model against various benchmarks in terms of object recognition, gaze behavior and adversarial robustness. Our findings suggest that the model can attend and gaze in ways similar to humans without being explicitly trained to mimic human attention, and that the model can enhance robustness against adversarial attacks due to its retinal sampling and recurrent processing. In particular, the model can correct its perceptual errors by taking more glances, setting itself apart from all feed-forward-only models. In conclusion, the interactions of retinal sampling, eye movement, and recurrent dynamics are important to human-like visual exploration and inference.
Human visual sensitivity to spatial details declines towards the periphery. Novel image synthesis techniques, so-called foveated rendering, exploit this observation and reduce the spatial resolution of synthesized images for the periphery, avoiding the synthesis of high-spatial-frequency details that are costly to generate but not perceived by a viewer. However, contemporary techniques do not make a clear distinction between the range of spatial frequencies that must be reproduced and those that can be omitted. For a given eccentricity, there is a range of frequencies that are detectable but not resolvable. While the accurate reproduction of these frequencies is not required, an observer can detect their absence if completely omitted. We use this observation to improve the performance of existing foveated rendering techniques. We demonstrate that this specific range of frequencies can be efficiently replaced with procedural noise whose parameters are carefully tuned to image content and human perception. Consequently, these frequencies do not have to be synthesized during rendering, allowing more aggressive foveation, and they can be replaced by noise generated in a less expensive post-processing step, leading to improved performance of the rendering system. Our main contribution is a perceptually-inspired technique for deriving the parameters of the noise required for the enhancement and its calibration. The method operates on rendering output and runs at rates exceeding 200FPS at 4K resolution, making it suitable for integration with real-time foveated rendering systems for VR and AR devices. We validate our results and compare them to the existing contrast enhancement technique in user experiments.
The requirements of much larger file sizes, different storage formats, and immersive viewing conditions of VR pose significant challenges to the goals of acquiring, transmitting, compressing, and displaying high-quality VR content. At the same time, the great potential of deep learning to advance progress on the video compression problem has driven a significant research effort. Because of the high bandwidth requirements of VR, there has also been significant interest in the use of space-variant, foveated compression protocols. We have integrated these techniques to create an end-to-end deep learning video compression framework. A feature of our new compression model is that it dispenses with the need for expensive search-based motion prediction computations. This is accomplished by exploiting statistical regularities inherent in video motion expressed by displaced frame differences. Foveation protocols are desirable since only a small portion of a video viewed in VR may be visible as a user gazes in any given direction. Moreover, even within a current field of view (FOV), the resolution of retinal neurons rapidly decreases with distance (eccentricity) from the projected point of gaze. In our learning based approach, we implement foveation by introducing a Foveation Generator Unit (FGU) that generates foveation masks which direct the allocation of bits, significantly increasing compression efficiency while making it possible to retain an impression of little to no additional visual loss given an appropriate viewing geometry. Our experiment results reveal that our new compression model, which we call the Foveated MOtionless VIdeo Codec (Foveated MOVI-Codec), is able to efficiently compress videos without computing motion, while outperforming foveated version of both H.264 and H.265 on the widely used UVG dataset and on the HEVC Standard Class B Test Sequences.
Virtual Reality is regaining attention due to recent advancements in hardware technology. Immersive images / videos are becoming widely adopted to carry omnidirectional visual information. However, due to the requirements for higher spatial and temporal resolution of real video data, immersive videos require significantly larger bandwidth consumption. To reduce stresses on bandwidth, foveated video compression is regaining popularity, whereby the space-variant spatial resolution of the retina is exploited. Towards advancing the progress of foveated video compression, we propose a full reference (FR) foveated image quality assessment algorithm, which we call foveated entropic differencing (FED), which employs the natural scene statistics of bandpass responses by applying differences of local entropies weighted by a foveation-based error sensitivity function. We evaluate the proposed algorithm by measuring the correlations of the predictions that FED makes against human judgements on the newly created 2D and 3D LIVE-FBT-FCVR databases for Virtual Reality (VR). The performance of the proposed algorithm yields state-of-the-art as compared with other existing full reference algorithms. Software for FED has been made available at: http://live.ece.utexas.edu/research/Quality/FED.zip
The spatially-varying field of the human visual system has recently received a resurgence of interest with the development of virtual reality (VR) and neural networks. The computational demands of high resolution rendering desired for VR can be offset by savings in the periphery, while neural networks trained with foveated input have shown perceptual gains in i.i.d and o.o.d generalization. In this paper, we present a technique that exploits the CUDA GPU architecture to efficiently generate Gaussian-based foveated images at high definition (1920x1080 px) in real-time (165 Hz), with a larger number of pooling regions than previous Gaussian-based foveation algorithms by several orders of magnitude, producing a smoothly foveated image that requires no further blending or stitching, and that can be well fit for any contrast sensitivity function. The approach described can be adapted from Gaussian blurring to any eccentricity-dependent image processing and our algorithm can meet demand for experimentation to evaluate the role of spatially-varying processing across biological and artificial agents, so that foveation can be added easily on top of existing systems rather than forcing their redesign (emulated foveated renderer). Altogether, this paper demonstrates how a GPU, with a CUDA block-wise architecture, can be employed for radially-variant rendering, with opportunities for more complex post-processing to ensure a metameric foveation scheme. Code is provided.
Segmentation of ultra-high resolution images is challenging because of their enormous size, consisting of millions or even billions of pixels. Typical solutions include dividing input images into patches of fixed size and/or down-sampling to meet memory constraints. Such operations incur information loss in the field-of-view (FoV) i.e., spatial coverage and the image resolution. The impact on segmentation performance is, however, as yet understudied. In this work, we start with a motivational experiment which demonstrates that the trade-off between FoV and resolution affects the segmentation performance on ultra-high resolution images---and furthermore, its influence also varies spatially according to the local patterns in different areas. We then introduce foveation module, a learnable "dataloader" which, for a given ultra-high resolution image, adaptively chooses the appropriate configuration (FoV/resolution trade-off) of the input patch to feed to the downstream segmentation model at each spatial location of the image. The foveation module is jointly trained with the segmentation network to maximise the task performance. We demonstrate on three publicly available high-resolution image datasets that the foveation module consistently improves segmentation performance over the cases trained with patches of fixed FoV/resolution trade-off. Our approach achieves the SoTA performance on the DeepGlobe aerial image dataset. On the Gleason2019 histopathology dataset, our model achieves better segmentation accuracy for the two most clinically important and ambiguous classes (Gleason Grade 3 and 4) than the top performers in the challenge by 13.1% and 7.5%, and improves on the average performance of 6 human experts by 6.5% and 7.5%. Our code and trained models are available at $\text{https://github.com/lxasqjc/Foveation-Segmentation}$.
The goal of this work is to characterize the representational impact that foveation operations have for machine vision systems, inspired by the foveated human visual system, which has higher acuity at the center of gaze and texture-like encoding in the periphery. To do so, we introduce models consisting of a first-stage \textit{fixed} image transform followed by a second-stage \textit{learnable} convolutional neural network, and we varied the first stage component. The primary model has a foveated-textural input stage, which we compare to a model with foveated-blurred input and a model with spatially-uniform blurred input (both matched for perceptual compression), and a final reference model with minimal input-based compression. We find that: 1) the foveated-texture model shows similar scene classification accuracy as the reference model despite its compressed input, with greater i.i.d. generalization than the other models; 2) the foveated-texture model has greater sensitivity to high-spatial frequency information and greater robustness to occlusion, w.r.t the comparison models; 3) both the foveated systems, show a stronger center image-bias relative to the spatially-uniform systems even with a weight sharing constraint. Critically, these results are preserved over different classical CNN architectures throughout their learning dynamics. Altogether, this suggests that foveation with peripheral texture-based computations yields an efficient, distinct, and robust representational format of scene information, and provides symbiotic computational insight into the representational consequences that texture-based peripheral encoding may have for processing in the human visual system, while also potentially inspiring the next generation of computer vision models via spatially-adaptive computation. Code + Data available here: https://github.com/ArturoDeza/EmergentProperties
In modern computer vision tasks, convolutional neural networks (CNNs) are indispensable for image classification tasks due to their efficiency and effectiveness. Part of their superiority compared to other architectures, comes from the fact that a single, local filter is shared across the entire image. However, there are scenarios where we may need to treat spatial locations in non-uniform manner. We see this in nature when considering how humans have evolved foveation to process different areas in their field of vision with varying levels of detail. In this paper we propose a way to enable CNNs to learn different pooling weights for each pixel location. We do so by introducing an extended definition of a pooling operator. This operator can learn a strict super-set of what can be learned by average pooling or convolutions. It has the benefit of being shared across feature maps and can be encouraged to be local or diffuse depending on the data. We show that for fixed network weights, our pooling operator can be computed in closed-form by spectral decomposition of matrices associated with class separability. Through experiments, we show that this operator benefits generalization for ResNets and CNNs on the CIFAR-10, CIFAR-100 and SVHN datasets and improves robustness to geometric corruptions and perturbations on the CIFAR-10-C and CIFAR-10-P test sets.
Emergent in the field of head mounted display design is a desire to leverage the limitations of the human visual system to reduce the computation, communication, and display workload in power and form-factor constrained systems. Fundamental to this reduced workload is the ability to match display resolution to the acuity of the human visual system, along with a resulting need to follow the gaze of the eye as it moves, a process referred to as foveation. A display that moves its content along with the eye may be called a Foveated Display, though this term is also commonly used to describe displays with non-uniform resolution that attempt to mimic human visual acuity. We therefore recommend a definition for the term Foveated Display that accepts both of these interpretations. Furthermore, we include a simplified model for human visual Acuity Distribution Functions (ADFs) at various levels of visual acuity, across wide fields of view and propose comparison of this ADF with the Resolution Distribution Function of a foveated display for evaluation of its resolution at a particular gaze direction. We also provide a taxonomy to allow the field to meaningfully compare and contrast various aspects of foveated displays in a display and optical technology-agnostic manner.
We consider the problem of fine-grained classification on an edge camera device that has limited power. The edge device must sparingly interact with the cloud to minimize communication bits to conserve power, and the cloud upon receiving the edge inputs returns a classification label. To deal with fine-grained classification, we adopt the perspective of sequential fixation with a foveated field-of-view to model cloud-edge interactions. We propose a novel deep reinforcement learning-based foveation model, DRIFT, that sequentially generates and recognizes mixed-acuity images.Training of DRIFT requires only image-level category labels and encourages fixations to contain task-relevant information, while maintaining data efficiency. Specifically, wetrain a foveation actor network with a novel Deep Deterministic Policy Gradient by Conditioned Critic and Coaching (DDPGC3) algorithm. In addition, we propose to shape the reward to provide informative feedback after each fixation to better guide RL training. We demonstrate the effectiveness of DRIFT on this task by evaluating on five fine-grained classification benchmark datasets, and show that the proposed approach achieves state-of-the-art performance with over 3X reduction in transmitted pixels.
This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.