Stereo rectification is widely considered “solved” due to the abundance of traditional approaches to perform recti-fication. However, autonomous vehicles and robots in-the-wild require constant re-calibration due to exposure to var-ious environmental factors, including vibration, and structural stress, when cameras are arranged in a wide-baseline configuration. Conventional rectification methods fail in these challenging scenarios: especially for larger vehicles, such as autonomous freight trucks and semi-trucks, the resulting incorrect rectification severely affects the quality of downstream tasks that use stereo/multi-view data. To tackle these challenges, we propose an online rectification approach that operates at real-time rates while achieving high accuracy. We propose a novel learning-based online cal-ibration approach that utilizes stereo correlation volumes built from a feature representation obtained from cross-image attention. Our model is trained to minimize vertical optical flow as proxy rectification constraint, and predicts the relative rotation between the stereo pair. The method is real-time and even outperforms conventional methods used for offline calibration, and substantially improves downstream stereo depth, post-rectification. We release two public datasets (https://light.princeton.edu/online-stereo-recification/), a synthetic and experimental wide baseline dataset, to foster further research.

Computer vision in unconstrained outdoor scenarios must tackle challenging high dynamic range (HDR) scenes and rapidly changing illumination conditions. Existing methods address this problem with multi-capture HDR sensors and a hardware image signal processor (ISP) that produces a single fused image as input to a downstream neural network. The output of the HDR sensor is a set of low dy-namic range (LDR) exposures, and the fusion in the ISP is performed in image space and typically optimized for hu-man perception on a display. Preferring tonemapped content with smooth transition regions over detail (and noise) in the resulting image, this image fusion does typically not preserve all information from the LDR exposures that may be essential for downstream computer vision tasks. In this work, we depart from conventional HDR image fusion and propose a learned task-driven fusion in the feature domain. Instead of using a single companded image, we introduce a novel local cross-attention fusion mechanism that exploits semantic features from all exposures learned in an end-to-end fashion with supervision from downstream detection losses. The proposed method outperforms all tested conventional HDR exposure fusion and auto-exposure methods in challenging automotive HDR scenarios.

Most camera lens systems are designed in isolation, separately from downstream computer vision methods. Recently, joint optimization approaches that design lenses alongside other components of the image acquisition and processing pipeline—notably, downstream neural networks—have achieved improved imaging quality or better performance on vision tasks. However, these existing methods optimize only a subset of lens parameters and cannot optimize glass materials given their categorical nature. In this work, we develop a differentiable spherical lens simulation model that accurately captures geometrical aberrations. We propose an optimization strategy to address the challenges of lens design—notorious for non-convex loss function landscapes and many manufacturing constraints—that are exacerbated in joint optimization tasks. Specifically, we introduce quantized continuous glass variables to facilitate the optimization and selection of glass materials in an end-to-end design context, and couple this with carefully designed constraints to support manufacturability. In automotive object detection, we report improved detection performance over existing designs even when simplifying designs to two- or three-element lenses, despite significantly degrading the image quality.

We propose Gated Stereo, a high-resolution and long-range depth estimation technique that operates on active gated stereo images. Using active and high dynamic range passive captures, Gated Stereo exploits multi-view cues alongside time-of-flight intensity cues from active gating. To this end, we propose a depth estimation method with a monocular and stereo depth prediction branch which are combined in a final fusion stage. Each block is supervised through a combination of supervised and gated self-supervision losses. To facilitate training and validation, we acquire a long-range synchronized gated stereo dataset for automotive scenarios. We find that the method achieves an improvement of more than 50 % MAE compared to the next best RGB stereo method, and 74 % MAE to existing monocular gated methods for distances up to 160 m. Our code, models and datasets are available here11https://light.princeton.edu/gatedstereo/.

Gated cameras hold promise as an alternative to scanning LiDAR sensors with high-resolution 3D depth that is robust to back-scatter in fog, snow, and rain. Instead of sequentially scanning a scene and directly recording depth via the photon time-of-flight, as in pulsed LiDAR sensors, gated imagers encode depth in the relative intensity of a handful of gated slices, captured at megapixel resolution. Although existing methods have shown that it is possible to decode high-resolution depth from such measurements, these methods require synchronized and calibrated LiDAR to supervise the gated depth decoder - prohibiting fast adoption across geographies, training on large unpaired datasets, and exploring alternative applications outside of automotive use cases. In this work, propose an entirely self-supervised depth estimation method that uses gated in-tensity profiles and temporal consistency as a training signal. The proposed model is trained end-to-end from gated video sequences, does not require LiDAR or RGB data, and learns to estimate absolute depth values. We take gated slices as input and disentangle the estimation of the scene albedo, depth, and ambient light, which are then used to learn to reconstruct the input slices through a cyclic loss. We rely on temporal consistency between a given frame and neighboring gated slices to estimate depth in regions with shadows and reflections. We experimentally validate that the proposed approach outperforms existing super-vised and self-supervised depth estimation methods based on monocular RGB and stereo images, as well as super-vised methods based on gated images. Code is available at https://github.com/princeton-computational-imaging/Gated2Gated.

Real-world scenes have a dynamic range of up to 280 dB that todays imaging sensors cannot directly capture. Existing live vision pipelines tackle this fundamental challenge by relying on high dynamic range (HDR) sensors that try to recover HDR images from multiple captures with different exposures. While HDR sensors substantially increase the dynamic range, they are not without disadvantages, including severe artifacts for dynamic scenes, reduced fill-factor, lower resolution, and high sensor cost. At the same time, traditional auto-exposure methods for low-dynamic range sensors have advanced as proprietary methods relying on image statistics separated from downstream vision algorithms. In this work, we revisit auto-exposure control as an alternative to HDR sensors. We propose a neural net-work for exposure selection that is trained jointly, end-to-end with an object detector and an image signal processing (ISP) pipeline. To this end, we use an HDR dataset for automotive object detection and an HDR training procedure. We validate that the proposed neural auto-exposure control, which is tailored to object detection, outperforms conventional auto-exposure methods by more than 6 points in mean average precision (mAP).

Recent implicit neural rendering methods have demonstrated that it is possible to learn accurate view synthesis for complex scenes by predicting their volumetric density and color supervised solely by a set of RGB images. However, existing methods are restricted to learning efficient representations of static scenes that encode all scene objects into a single neural network, and they lack the ability to represent dynamic scenes and decompose scenes into individual objects. In this work, we present the first neural rendering method that represents multi-object dynamic scenes as scene graphs. We propose a learned scene graph representation, which encodes object transformations and radiance, allowing us to efficiently render novel arrangements and views of the scene. To this end, we learn implicitly encoded scenes, combined with a jointly learned latent representation to describe similar objects with a single implicit function. We assess the proposed method on synthetic and real automotive data, validating that our approach learns dynamic scenes – only by observing a video of this scene – and allows for rendering novel photo-realistic views of novel scene compositions with unseen sets of objects at unseen poses.

Adversarial attacks play a critical role in understanding deep neural network predictions and improving their robustness. Existing attack methods aim to deceive convolutional neural network (CNN)-based classifiers by manipulating RGB images that are fed directly to the classifiers. However, these approaches typically neglect the influence of the camera optics and image processing pipeline (ISP) that produce the network inputs. ISPs transform RAW measurements to RGB images and traditionally are assumed to preserve adversarial patterns. In fact, these low-level pipelines can destroy, introduce or amplify adversarial patterns that can deceive a downstream detector. As a result, optimized patterns can become adversarial for the classifier after being transformed by a certain camera ISP or optical lens system but not for others. In this work, we examine and develop such an attack that deceives a specific camera ISP while leaving others intact, using the same downstream classifier. We frame this camera-specific attack as a multi-task optimization problem, relying on a differentiable approximation for the ISP itself. We validate the proposed method using recent state-of-the-art automotive hardware ISPs, achieving 92% fooling rate when attacking a specific ISP. We demonstrate physical optics attacks with 90% fooling rate for a specific camera lens.

Today’s state-of-the-art methods for 3D object detection are based on lidar, stereo, or monocular cameras. Lidar-based methods achieve the best accuracy, but have a large footprint, high cost, and mechanically-limited angular sampling rates, resulting in low spatial resolution at long ranges. Recent approaches using low-cost monocular or stereo cameras promise to overcome these limitations but struggle in low-light or low-contrast regions as they rely on passive CMOS sensors. We propose a novel 3D object detection modality that exploits temporal illumination cues from a low-cost monocular gated imager. We introduce a novel deep detection architecture, Gated3D, that is tailored to temporal illumination cues in gated images. This modality allows us to exploit mature 2D object feature extractors that guide the 3D predictions through a frustum segment estimation. We assess the proposed method experimentally on a 3D detection dataset that includes gated images captured over 10,000 km of driving data. We validate that our method outperforms state-of-the-art monocular and stereo methods, opening up a new sensor modality as an avenue to replace lidar in autonomous driving. https://light.princeton.edu/gated3d

Most modern commodity imaging systems we use directly for photography—or indirectly rely on for downstream applications—employ optical systems of multiple lenses that must balance deviations from perfect optics, manufacturing constraints, tolerances, cost, and footprint. Although optical designs often have complex interactions with downstream image processing or analysis tasks, today’s compound optics are designed in isolation from these interactions. Existing optical design tools aim to minimize optical aberrations, such as deviations from Gauss’ linear model of optics, instead of application-specific losses, precluding joint optimization with hardware image signal processing (ISP) and highly parameterized neural network processing. In this article, we propose an optimization method for compound optics that lifts these limitations. We optimize entire lens systems jointly with hardware and software image processing pipelines, downstream neural network processing, and application-specific end-to-end losses. To this end, we propose a learned, differentiable forward model for compound optics and an alternating proximal optimization method that handles function compositions with highly varying parameter dimensions for optics, hardware ISP, and neural nets. Our method integrates seamlessly atop existing optical design tools, such as Zemax . We can thus assess our method across many camera system designs and end-to-end applications. We validate our approach in an automotive camera optics setting—together with hardware ISP post processing and detection—outperforming classical optics designs for automotive object detection and traffic light state detection. For human viewing tasks, we optimize optics and processing pipelines for dynamic outdoor scenarios and dynamic low-light imaging. We outperform existing compartmentalized design or fine-tuning methods qualitatively and quantitatively, across all domain-specific applications tested.

Commodity imaging systems rely on hardware image signal processing (ISP) pipelines. These low-level pipelines consist of a sequence of processing blocks that, depending on their hyperparameters, reconstruct a color image from RAW sensor measurements. Hardware ISP hyperparameters have a complex interaction with the output image, and therefore with the downstream application ingesting these images. Traditionally, ISPs are manually tuned in isolation by imaging experts without an end-to-end objective. Very recently, ISPs have been optimized with 1st-order methods that require differentiable approximations of the hardware ISP. Departing from such approximations, we present a hardware-in-the-loop method that directly optimizes hardware image processing pipelines for end-to-end domain-specific losses by solving a nonlinear multi-objective optimization problem with a novel 0th-order stochastic solver directly interfaced with the hardware ISP. We validate the proposed method with recent hardware ISPs and 2D object detection, segmentation, and human viewing as end-to-end downstream tasks. For automotive 2D object detection, the proposed method outperforms manual expert tuning by 30% mean average precision (mAP) and recent methods using ISP approximations by 18% mAP.

The fusion of multimodal sensor streams, such as camera, lidar, and radar measurements, plays a critical role in object detection for autonomous vehicles, which base their decision making on these inputs. While existing methods exploit redundant information in good environmental conditions, they fail in adverse weather where the sensory streams can be asymmetrically distorted. These rare ``edge-case'' scenarios are not represented in available datasets, and existing fusion architectures are not designed to handle them. To address this challenge we present a novel multimodal dataset acquired in over 10,000~km of driving in northern Europe. Although this dataset is the first large multimodal dataset in adverse weather, with 100k labels for lidar, camera, radar, and gated NIR sensors, it does not facilitate training as extreme weather is rare. To this end, we present a deep fusion network for robust fusion without a large corpus of labeled training data covering all asymmetric distortions. Departing from proposal-level fusion, we propose a single-shot model that adaptively fuses features, driven by measurement entropy. We validate the proposed method, trained on clean data, on our extensive validation dataset. Code and data are available here https://github.com/princeton-computational-imaging/SeeingThroughFog.

Conventional sensor systems record information about directly visible objects, whereas occluded scene components are considered lost in the measurement process. Non-line-of-sight (NLOS) methods try to recover such hidden objects from their indirect reflections - faint signal components, traditionally treated as measurement noise. Existing NLOS approaches struggle to record these low-signal components outside the lab, and do not scale to large-scale outdoor scenes and high-speed motion, typical in automotive scenarios. In particular, optical NLOS capture is fundamentally limited by the quartic intensity falloff of diffuse indirect reflections. In this work, we depart from visible-wavelength approaches and demonstrate detection, classification, and tracking of hidden objects in large-scale dynamic environments using Doppler radars that can be manufactured at low-cost in series production. To untangle noisy indirect and direct reflections, we learn from temporal sequences of Doppler velocity and position measurements, which we fuse in a joint NLOS detection and tracking network over time. We validate the approach on in-the-wild automotive scenes, including sequences of parked cars or house facades as relay surfaces, and demonstrate low-cost, real-time NLOS in dynamic automotive environments.

Conventional intensity cameras recover objects in the direct line-of-sight of the camera, whereas occluded scene parts are considered lost in this process. Non-line-of-sight imaging (NLOS) aims at recovering these occluded objects by analyzing their indirect reflections on visible scene surfaces. Existing NLOS methods temporally probe the indirect light transport to unmix light paths based on their travel time, which mandates specialized instrumentation that suffers from low photon efficiency, high cost, and mechanical scanning. We depart from temporal probing and demonstrate steady-state NLOS imaging using conventional intensity sensors and continuous illumination. Instead of assuming perfectly isotropic scattering, the proposed method exploits directionality in the hidden surface reflectance, resulting in (small) spatial variation of their indirect reflections for varying illumination. To tackle the shape-dependence of these variations, we propose a trainable architecture which learns to map diffuse indirect reflections to scene reflectance using only synthetic training data. Relying on consumer color image sensors, with high fill factor, high quantum efficiency and low read-out noise, we demonstrate high-fidelity color NLOS imaging for scene configurations tackled before with picosecond time resolution.

Nearly every commodity imaging system we directly interact with, or indirectly rely on, leverages power efficient, application-adjustable black-box hardware image signal processing (ISPs) units, running either in dedicated hardware blocks, or as proprietary software modules on programmable hardware. The configuration parameters of these black-box ISPs often have complex interactions with the output image, and must be adjusted prior to deployment according to application-specific quality and performance metrics. Today, this search is commonly performed manually by "golden eye" experts or algorithm developers leveraging domain expertise. We present a fully automatic system to optimize the parameters of black-box hardware and software image processing pipelines according to any arbitrary (i.e., application-specific) metric. We leverage a differentiable mapping between the configuration space and evaluation metrics, parameterized by a convolutional neural network that we train in an end-to-end fashion with imaging hardware in-the-loop. Unlike prior art, our differentiable proxies allow for high-dimension parameter search with stochastic first-order optimizers, without explicitly modeling any lower-level image processing transformations. As such, we can efficiently optimize black-box image processing pipelines for a variety of imaging applications, reducing application-specific configuration times from months to hours. Our optimization method is fully automatic, even with black-box hardware in the loop. We validate our method on experimental data for real-time display applications, object detection, and extreme low-light imaging. The proposed approach outperforms manual search qualitatively and quantitatively for all domain-specific applications tested. When applied to traditional denoisers, we demonstrate that---just by changing hyperparameters---traditional algorithms can outperform recent deep learning methods by a substantial margin on recent benchmarks.

Depth from defocus (DFD) requires estimating the depth dependent defocus blur at every pixel. Several approaches for accomplishing this have been proposed over the years. For a pair of images this is done by modeling the defocus relationship between the two differently defocused images and for single defocused images by relying on the the properties of the point spread function and the characteristics of the latent sharp image. We propose depth discriminative filters for DFD that can represent many of the widely used models such as the relative blur, Blur Equalization Technique, deconvolution based depth estimation, and subspace projection methods. We show that by optimizing the parameters of this general model we can obtain state-of-the-art result on synthetic and real defocused images with single or multiple defocused images with different apertures.

Pictures taken with finite aperture lenses typically have out-of-focus regions. While such defocus blur is useful for creating photographic effects, it can also be used for depth estimation. In this paper, we look at different camera settings for Depth from Defocus (DFD), the conditions under which depth can be estimated unambiguously for those settings and optimality of different settings in terms of lower bound of error variance. We present results for general camera settings, as well as two of the most widely used camera settings namely, variable aperture and variable focus. We show that for variable focus, the range of depth needs to be larger than twice the focal length to unambiguously estimate depth. We analytically derive the optimal aperture ratio, and also show that there is no single optimal parameter for variable focus. Furthermore we show how to choose focus in order to minimize error variance in a particular region of the scene.