Though Deep Neural Networks (DNN) show excellent performance across various computer vision tasks, several works show their vulnerability to adversarial samples, i.e., image samples with imperceptible noise engineered to manipulate the network's prediction. Adversarial sample generation methods range from simple to complex optimization techniques. Majority of these methods generate adversaries through optimization objectives that are tied to the pre-softmax or softmax output of the network. In this work we, (i) show the drawbacks of such attacks, (ii) propose two new evaluation metrics: Old Label New Rank (OLNR) and New Label Old Rank (NLOR) in order to quantify the extent of damage made by an attack, and (iii) propose a new adversarial attack FDA: Feature Disruptive Attack, to address the drawbacks of existing attacks. FDA works by generating image perturbation that disrupt features at each layer of the network and causes deep-features to be highly corrupt. This allows FDA adversaries to severely reduce the performance of deep networks. We experimentally validate that FDA generates stronger adversaries than other state-of-theart methods for image classification, even in the presence of various defense measures. More importantly, we show that FDA disrupts feature-representation based tasks even without access to the task-specific network or methodology. 1
Metasurfaces have the potential to revolutionize imaging technologies due to their extreme control of phase, polarization, and amplitude of the incident light. They rely upon enhanced local interaction of light to achieve the desired phase profile. As a consequence of the enhanced local interaction of light, metasurfaces are highly dispersive. This strong dispersion has been recognized as a primary limitation as it relates to realizing conventional imaging with metasurfaces. Here, we argue that this strong dispersion is an added degree of design freedom for computational imaging, potentially opening up novel applications. In particular, we exploit this strongly dispersive property of metasurfaces to propose a compact, single-shot, and passive 3D imaging camera. Our device consists of a metalens engineered to focus different wavelengths at different depths and two deep networks to recover depth and RGB texture information from chromatic, defocused images acquired by the system. In contrast with other metasurface-based 3D sensors, our design can operate in the full visible range with a larger field-of-view (FOV) and can potentially generate dense depth maps of complicated 3D scenes. Our simulation results on a 1 mm diameter metalens demonstrate its ability to capture 3D depth and texture information ranging from 0.12 to 0.6 m.
Lensless imaging provides opportunities to design imaging systems free from the constraints imposed by traditional camera architectures. Due to advances in imaging hardware, fabrication techniques, and new algorithms, researchers have recently developed lensless imaging systems that are extremely compact and lightweight or able to image higher-dimensional quantities. Here we review these recent advances and describe the design principles and their effects that one should consider when developing and using lensless imaging systems.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.