Excel to HTML
MONDAY 2 MARCH 2026
Superresolution
Session Chair: Emma Reid, Oak Ridge National Laboratory
08:30 - 10:30
Grand Peninsula C
08:30COIMG-121
Deep learning-based super-resolution for X-ray computed tomography in additive manufacturing, Obaidullah Rahman, Oak Ridge National Laboratory (US); Haley Duba-Sullivan, Oak Ridge National Laboratory (US); Amirkoushyar Ziabari, Oak Ridge National Laboratory (US) [view abstract]
Industrial X-ray computed tomography (XCT) is a non-destructive method for inspection and characterization of additively manufactured (AM) components. XCT resolution is limited by detector binning, field-of-view restriction, system blur, motion, and acquisition settings. These limitations reduce the detectability of critical flaws. Super-resolution (SR) techniques offer a promising solution for improving the effective resolution and image quality without any expensive hardware upgrades or time-consuming scans. Deep learning (DL)-based SR methods are a powerful tool for reconstructing high-resolution volumes from low-resolution inputs.In this work, a novel DL-based SR method is proposed for XCT scans of AM. The proposed method, Simurgh-SR, is built on the pre-existing Simurgh framework and consists of a 2.5D U-Net trained to map low-quality inputs to high-quality reconstructions. Simurgh-SR demonstrates superior performance in 4X SR on real XCT scans of thick 316L AM components, enhancing the structural similarity score and peak signal-to-noise ratio (>8 dB). We also compared against the state-of-the-art (ESRGAN) and demonstrated superior performance in flaw detection (>2.3X). This enables more accurate and faster characterization of AM components. Simurgh-SR, trained for both 2X and 4X SR, performs effectively at both levels, enabling the use of a single model for various SR factors.
08:50COIMG-123
Fixed-basis low-rank tensor approximation for the fusion of hyperspectral and multispectral imagery, James Fowler, Mississippi State University (US) [view abstract]
The inverse-imaging problem comprising the fusing of a hyperspectral image, possessing high spectral resolution, with a multispectral image, having high spatial resolution, to yield an image with high resolution both spatially and spectrally is considered. In particular, a prior state-of-the-art approach low-rank tensor approximation (LRTA) is revisited with the goal of simplifying its implementation and accelerating its execution speed. Whereas the original LRTA incorporated low-rank objectives both spatially and spectrally, the revised algorithm employs spectral low-rankness exclusively. Additionally, the reliance of LRTA on singular value thresholding (SVT) an operator widely used to impose low-rankness in optimizations is replaced with a fixed-basis approximation that eliminates the computationally costly singular value decomposition required by the SVT. The proposed modifications ultimately result in significant runtime speedup; furthermore, empirical results reveal improved fusion quality when compared to the original LRTA.
09:10COIMG-122
Constrained conditional denoising diffusion for hyperspectral-multispectral fusion, James Fowler, Mississippi State University (US) [view abstract]
Diffusion models have recently risen to prominence for a variety of inverse-imaging problems. Many such models use what is commonly known as condition diffusion which effectively samples from the distribution of the desired image conditioned on some known side information, often a degraded or lower-resolution version of the target image.> However, an alternative paradigm has recently emerged in the form of constrained diffusion in which explicit constraints between the target image and the side information are iteratively incorporated into the diffusion reconstruction during inference.> While prior literature has considered conditional and constrained diffusion to effectively be mutually exclusive, a diffusion algorithm is proposed to combine the two within the widely-used denoising diffusion probabilistic models (DDPM) framework.> The resulting approach constrained conditional denoising diffusion inputs both the target and side information into the diffusion network during both training and inference similar to conditional diffusion but also applies explicit constraints during inference like constrained diffusion.> The proposed approach is evaluated for the task of fusing a hyperspectral image, possessing high spectral resolution, with a multispectral image, having high spatial resolution, to yield an image with high resolution both spatially and spectrally, an inverse-imaging problem called hyperspectral-multispectral fusion.> Experimental results demonstrate that, not only can constrained and conditional diffusion operate complementarily and achieve performance superior to either used alone, but also that the proposed constrained conditional denoising diffusion outperforms other state-of-the-art approaches for hyperspectral-multispectral fusion.
09:30COIMG-119
Shedding light on the night: Multi-modal super-resolution for NTL imagery, Haley Sullivan, Oak Ridge National Lab; Tony Allen, ; Emma Reid, [view abstract]
Nighttime light (NTL) imagery provides valuable insights into urbanization, disaster response, and energy consumption. However, there is an inherent trade-off between revisit rate and spatial resolution: while the Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS DNB) delivers daily global coverage needed for timely analysis, its coarse resolution limits the ability to do detailed assessments. Super-resolution techniques offer a way to enhance spatial detail, enabling improved assessments of infrastructure, light pollution, economic activity, and power outages. Yet, existing state-of-the-art methods are not optimized for the unique properties of NTL data. This work introduces a multi-modal super-resolution framework that leverages the strong correlation between NTL images, road networks, and land use maps. By integrating VIIRS DNB imagery with ancillary geographic features, our approach improves reconstruction accuracy and yields higher-fidelity spatial detail over standard super-resolution methods. We demonstrate the effectiveness of our method on real-world NTL datasets, showing significant improvements in both visual quality and quantitative metrics.
09:50COIMG-120
The double-edged sword of data-driven super-resolution: Adversarial super-resolution models, Haley Duba-Sullivan, Oak Ridge National Laboratory; Purdue University (US); Steven Young, Oak Ridge National Laboratory; Emma Reid, Oak Ridge National Laboratory [view abstract]
Data-driven super-resolution methods are essential components in imaging pipelines, but their reliance on learned representations introduces unexpected vulnerabilities. We present AdvSR, a framework demonstrating that adversarial behavior can be embedded directly into SR model weights during training, requiring no access to inputs at inference time. AdvSR jointly optimizes for reconstruction quality and targeted adversarial outcomes, producing models that appear benign while inducing downstream misclassification. This work highlights a critical but unexplored dimension of data-driven SR: the training data and optimization process themselves can become attack vectors, with implications for how we build and deploy these models.
Wave-Based Scientific Imaging (Joint Session with Machine Learning for Scientific Imaging)
15:30 - 16:30
Grand Peninsula C
15:30MLSI-301
Coherent computational imaging with synthetic waves, Florian Willomitzer, University of Arizona (US) [view abstract]
Imaging through scattering scenes or materials severely limits the visual acuity of optical imaging systems. This talk discusses how diversity in illumination wavelength can be utilized to circumvent the problem of phase randomization in scattered light fields. Our technique probes the scene at two closely spaced optical wavelengths and computationally assembles a complex "synthetic field" at a "synthetic wavelength," which is used for further processing. As the synthetic wavelength is the beat wavelength of the two used optical "carrier" wavelengths, it can be picked orders of magnitudes larger, and the computationally assembled synthetic field becomes immune to scatter. Amongst other applications, the introduced method allows for holographic measurements of hidden objects through scattering media or around corners, or for interferometric measurements of macroscopic objects with rough surfaces. During the talk, different flavors of the technique will be introduced, including a method to retrieve the complex synthetic field in single-shot.
15:50MLSI-302
Generative AI and pre-trained foundation models for seismic imaging and interpretation, Nori Nakata, Berkeley Lab (US) [view abstract]
Recent advances in Generative AI (GenAI) and pre-trained general-purpose foundation models are redefining how we image and interpret the subsurface. We present an integrated framework that unifies physics-informed generative modeling, diffusion-based synthesis, and transfer-learned foundation models for seismic imaging and interpretation. Generative diffusion transformers simulate broadband seismic wavefields consistent with elastic physics, enabling virtual dense-array simulations and uncertainty-aware data augmentation. Building on these physics-consistent representations, pre-trained vision foundation models are adapted to the seismic domain via parameter-efficient fine-tuning--using low-rank adaptation and prefix prompting--to retain general visual priors while embedding geological constraints through relative-geologic-time and graph-based reasoning. Self-supervised neural operators coupled with partial-differential-equation constraints unfold and unfault complex 3-D structures, allowing geological interpretation of seismic images without significant human expertise. Together, these components establish a scalable GenAI-foundation-model framework that bridges physical modeling, representation learning, and structural reasoning. By unifying pre-trained representations and geological priors, this paradigm moves seismic interpretation beyond task-specific networks toward adaptive, interpretable, and foundation-scale digital twins of the subsurface--marking a new era for data-driven Earth imaging.
16:10MLSI-303
A CNN workflow for stochastic seismic property estimation, Haibo Di, SLB (US) [view abstract]
As one of the major tools in resolving the non-uniqueness challenge in subsurface interpretation and reservoir characterization, stochastic inversion from post- and pre-stack seismic data remains challenging, which not only requires heavy computational resources but also relies on intensive manual supervision. Inspired by the recent advances in deep learning particularly convolutional neural networks (CNNs) for interdisciplinary data integration, in this talk, we will first present a deep learning workflow that enables stochastic property estimation by efficiently integrating seismic images with sparse wells. It starts with sampling a set of property prior models (PPMs) from densely-measured actual logs at well locations and corrupting local seismic patterns with Gaussian noise. The core idea is to train a structure-guided CNN by mapping the contaminated seismic with the sampled PPMs while enforcing structural consistency to avoid overfitting in the common scenario of sparse wells. The baseline and uncertainty of target properties are estimated by running multiple realizations of the trained CNN. Next, we will demonstrate how the proposed workflow performs over three use cases, including rock acoustic/elastic property estimation from 3D post-/angle-stack seismic and soil geotechnical properties from 2D ultra-high-resolution seismic. Finally, we end with a summary and discussion on future work.
Computational Imaging: Why Hire?
Session Chair: Stanley Chan, Purdue University
16:30 - 17:30
Grand Peninsula C
16:30COIMG-124
Value, the value of your value and how to monetize it - engineering a successful and rewarding career, Sergiu Goma, Alkemista Tech [view abstract]
This talk will discuss value as a unique human outcome, how value is validated and evaluated in a specialized industrial society and how to plan and design a successful career in engineering. How value predicates money and the role and scope of currency in engineering.
16:50COIMG-125
Translating PhD brilliance into industry impact: Crucial skills in the AI era, Ying Chen Lou, Google [view abstract]
TBD
17:10COIMG-126
The new imaging stack: Essential skills for the generative AI era, Hossein Talebi, Google Inc. [view abstract]
As computational photography evolves from classical signal processing to deep learning and generative models, the profile of a successful graduate hire is changing rapidly. In this talk, I will discuss the critical skills required to build modern imaging products, moving beyond standard model design to include optimization, data pipeline management, and latency-constrained inference. We will explore how students can better prepare for industry by balancing fundamental knowledge of imaging with the practical software engineering needed to deploy large-scale generative AI models for consumer applications.
17:30COIMG-Panel2
Finding Jobs in the AI Singularity, Stanley Chan, Purdue University (US) [view abstract]
Rapid advances in AI are reshaping research, industry, and the academic–industrial boundary, often faster than traditional career pathways can adapt. This panel brings together faculty, industry researchers, startup leaders, and recent graduates to discuss how computational imaging researchers, PhD students, and postdocs can navigate careers in an era of accelerating AI capability. The discussion aims to provide practical guidance, realistic perspectives, and strategic insight for early- and mid-career researchers. Panelists: Ying Chen Lou (Google Pixel Camera), Stanley Chan (Purdue University), Sergio Goma (Sinagua Malt), Hossein Talebi (Google Research).
TUESDAY 3 MARCH 2026
Generative AI for Imaging (Joint Session with Machine Learning for Scientific Imaging)
08:30 - 09:30
Grand Peninsula C
08:30MLSI-304
Metric 3D human pose estimation via CAD-calibrated monocular depth, Sakthivel Sivaraman, Nvidia (US) [view abstract]
The estimation of 3D human pose from a single 2D image is a fundamentally ill-posed problem due to the inherent ambiguity between an object's scale and its depth. This challenge is particularly pronounced in applications within geometrically constrained environments, where metric-scale (absolute) joint positions are critical for real-world interaction, monitoring, or safety systems. This work presents a novel, multi-model deep learning framework that achieves robust absolute 3D pose estimation using only a single RGB or IR camera. By decoupling the tasks of relative pose estimation and absolute depth inference and introducing environment-specific geometric priors, our system overcomes the limitations of traditional monocular approaches. Our work uses a vehicle cabin as a primary example of such a constrained environment for in-cabin sensing, but the methodology is applicable to robotics, surveillance, factory automation etc.
08:50MLSI-305
Self-supervised learning for spatiotemporally scalable computational imaging, Kevin Zhou, University of Michigan (US) [view abstract]
Spatiotemporal throughput can be defined as the number of pixels or voxels per second an imaging system can capture, governing its field of view, spatial resolution, and frame/volume rate. Conventional optical imaging systems face a variety of bottlenecks and tradeoffs that limit their spatiotemporal throughput, such as scanning, asynchronous/sequential acquisitions, aberrations, and limited detector bandwidths. This talk presents multiple large-scale computational imaging systems that alleviate such bottlenecks through a synchronized array of cameras and a corresponding large-scale reconstruction algorithm that computationally fuses the parallel video streams. While such parallelized computational imaging systems can capture data at very high throughput, the associated iterative reconstruction algorithms struggle to keep up, especially as the number of cameras and/or the length of the video increases. To this end, we discuss self-supervised learning-based strategies that establish a mapping between the raw data and the desired reconstruction using a fraction of the acquired data. The trained model can then generalize to the remainder/future unseen frames, enabling scalability to arbitrarily long videos.
Imaging Pipelines I (Joint Session with Machine Learning for Scientific Imaging)
09:30 - 10:30
Grand Peninsula C
09:30COIMG-127
Computational imaging for aberration correction, Laura Waller, UC Berkeley (US) [view abstract]
Computational imaging is permeating cameras and microscopes across many scientific applications, enabling new high-resolution and multi-dimensional measurement capabilities (e.g. phase, 3D, hyperspectral). Many methods require acquisition of multiple images across a large field-of-view, so motion blur and aberrations can cause severe artifacts. This talk will describe new space-time algorithms that correct aberrations and motion artifacts with imperfect optical systems or approximate forward models. Traditional model-based image reconstruction algorithms work together with neural networks to optimize the inverse problem solver and the data capture strategy.
09:50COIMG-128
Efficient-coding--inspired in-sensor compression, Yuhao Zhu, University of Rochester (US) [view abstract]
Energy-efficient image acquisition on the edge is crucial for enabling remote sensing applications where the sensor node has weak compute capabilities and must transmit data to a remote server/cloud for processing. To reduce the edge energy consumption, this paper proposes a sensor-algorithm co-designed system, which compresses raw pixels in the analog domain inside the sensor. We use coded exposure (CE) as the in-sensor compression strategy as it offers the flexibility to sample, i.e., selectively expose pixels, both spatially and temporally. We make three contributions. First, we propose a task-agnostic strategy to learn the sampling/exposure pattern based on the classic theory of efficient coding. Second, we co-design the downstream vision model with the exposure pattern to address the pixel-level non-uniformity unique to CE-compressed images. Finally, we propose lightweight augmentations to the image sensor hardware to support our in-sensor CE compression. Evaluating on action recognition and video reconstruction, our system outperforms state-of-the-art video-based methods at the same speed while reducing the energy by up to 15.4 times.
Multi-Modal Imaging
Session Chair: Mel White, Rice University
11:00 - 12:20
Grand Peninsula C
11:00COIMG-130
Fusing optics and acoustics: From data integration to physics-level modulation, Adithya Pediredla, Dartmouth University (US) [view abstract]
Multimodal fusion traditionally refers to capturing a scene with different sensing modalities such as optical and acoustic and combining their outputs through algorithmic or learning-based techniques to obtain more detailed reconstructions. These approaches have enabled faster and more comprehensive small-baseline reconstructions, particularly with recent advances in neural scene representations. In this talk, I will first present neural scene representations that unify diverse imaging modalities, enabling the reconstruction of dynamic 3D scenes at small baselines and high speed, thereby expanding the ability to image complex environments. Alongside this, I will also present a new form of multimodal fusion that operates at a deeper level: rather than combining modalities after capture, we use sound at the physics level to directly modulate light as it is being measured by the imaging system. By embedding acoustic optical interactions into the hardware and sensing process itself, we create fundamentally new imaging systems that can steer and structure light at megahertz rates orders of magnitude faster than mechanical methods. This physical fusion, combined with modern processing and neural rendering frameworks, enables ultra-fast, robust 3D reconstruction.
11:20COIMG-131
Spatial reasoning: Can AI understand the 3D world?, Achuta Kadambi, UCLA (US) [view abstract]
Vision-language models (VLMs) have achieved impressive progress in combining visual understanding with linguistic reasoning. Yet, they remain fundamentally limited when it comes to interpreting dynamic spatiotemporal interactions. This talk discusses the state of the art in VLM reasoning, and to what extent they understand the spatial world around us.
11:40COIMG-132
Unified imaging approaches with event-based and frame-based sensors, Joseph Cox, Air Force Research Laboratory (US) [view abstract]
Event-based sensors (EBS) are imaging sensors equipped with an irradiance change detection read-out circuit at each pixel. By generating output only when a significant change is detected, EBS have been practically shown to reduce read-out bandwidth and processing requirements by orders of magnitude compared to frame-based sensors. However, there are potential disadvantages, such as EBS reading out events corresponding to the edges of moving objects, leading to degradation in effective image quality.This talk explores the usage of EBS for potential defense applications. First, we review our past experiments and public literature to understand the costs and benefits of EBS. These experiments include bandwidth measurement, object detection task performance, optical contrast amplification, and an initial hybrid event/frame-based object tracking system. Based on this understanding, we discuss our research direction focused on augmenting existing, proven frame-based sensor applications with event-based sensors. Finally, we present new experiments in developing hybrid tracking systems. Listeners to this talk will gain an understanding of EBS and measurement of its costs and benefits, potential design approaches with the technology, and examples of its application.
12:00COIMG-173
Multi-modal computational 3D imaging for specular surfaces, Jiazhang Wang, University of Arizona (US); Florian Willomitzer, [view abstract]
Despite its importance in industrial quality control, AR/VR, or medical imaging, the robust and accurate 3D measurement of specular and mixed reflectance surfaces continues to be a key unmet need for current state-of-the-art approaches.This talk presents a series of metrology-inspired techniques that leverage efficient illumination encoding combined with additional imaging modalities, such as polarization, for high-quality computational 3D imaging on specular surfaces. I will mainly introduce a novel multi-modal method that fuses polarimetric and deflectometric information. Both physical and data-driven evaluation strategies will be discussed, alongside their fundamental limits and trade-offs.
Semiconductor Metrology and Computational Lithography (Joint Session with Machine Learning for Scientific Imaging)
15:30 - 17:30
Grand Peninsula C
15:30COIMG-137
It's just flat: Optimization landscape of inverse lithography, Stanley Chan, Purdue University (US) [view abstract]
Many people characterize the optimization in inverse lithography as nonlinear, nonconvex, nonsmooth, hard-to-converge, ill-posed, many local minima, etc, which essentially portraits the subject as one of the evilest combinations of the worst adjectives in optimization textbooks. But is the optimization landscape of inverse lithography really this horrible, aka with peaks and valleys, and with spikes everywhere plus a lot of singularities? We took a closer look at the optimization landscape of the problem. We found that there isn't anything complicated about the optimization landscape. It is just flat. We will explain the source of this flatness, and techniques to overcome the flatness.
16:50COIMG-138
Advancing non-destructive evaluation in electronics: 3D X-ray microscopy and deep learning for failure analysis and quality assurance, Herminso Villarraga-Gomez, Zeiss (US) [view abstract]
X-ray-based technologies, initially developed for medical applications, have become essential tools for non-destructive evaluation (NDE) in the electronics industry. Techniques such as radiography, computed tomography (CT), and 3D X-ray microscopy (XRM) enable detailed assembly analysis, structural characterization, and fault isolation in electronic devices, including printed circuit boards (PCBs), system-in-package technologies, and advanced packaging architectures. While traditional 2D radiography and laminography provide valuable insights, they face limitations in resolving overlapping materials and obstructed components. Advanced 3D XRM systems, featuring resolution-at-a-distance (RaaD) capabilities and deep learning (DL)-based reconstruction algorithms, overcome these challenges by delivering high-resolution imaging of sub-micrometer and nanoscale features. DL-based algorithms, such as DeepRecon and DeepScout, significantly enhance image quality, reduce noise, and improve data acquisition efficiency by up to 10x. These advancements enable faster, more accurate failure analysis (FA) and quality assurance workflows, while preserving the integrity of electronic components. Furthermore, correlative imaging workflows, integrating 3D XRM with techniques like scanning electron microscopy (SEM) and transmission electron microscopy (TEM), provide multiscale insights into complex electronic structures. By addressing the challenges of modern electronics manufacturing, 3D X-ray imaging and DL-based innovations support the development of reliable, high-performance devices, meeting the growing demands of the global electronics industry.
17:10COIMG-166
Fast X-ray micro- and nano-laminography imaging of integrated circuits at the upgraded Advanced Photon Source, Viktor Nikitin, Argonne National Laboratory (US) [view abstract]
WEDNESDAY 4 MARCH 2026
Wavefront Estimation and Imaging I
08:30 - 10:30
Grand Peninsula C
08:30COIMG-140
ReVAR: A statistical simulation algorithm for aero-optic phase screens, Jeffrey Utley, Purdue University (US); Gregery Buzzard, Purdue University (US); Charles Bouman, Purdue University (US); Matthew Kemnetz, Air Force Institute of Technology (US) [view abstract]
Abstract pending OpSec approval
08:50COIMG-141
Wind tunnel tomographic wavefront sensing simulations, Karl Weisenburger, Purdue University (US); Gregery Buzzard, Purdue University (US); Charles Bouman, Purdue University (US); Matthew Kemnetz, Air Force Institute of Technology (US) [view abstract]
Abstract pending OpSec approval.
09:10COIMG-142
Split-step phase retrieval: Recovering layered phase screens, Nicholas Chimitt, Purdue University (US) [view abstract]
A wave propagated through layered media, such as atmospheric turbulence, incurs degradation and limits the optical resolution of imaging devices. The state of the layered media and its recovery is important to applications related to channel characterization, advanced adaptive optics systems, and beam pre-compensation. Typically, the recovery of phase-based scattering screens is achieved through (i) measuring the incident wavefront directly using wavefront sensors or (ii) using holographic measurements. We demonstrate that it is possible to recover phase screens using intensity measurements of point sources propagated through the layered phase screens, combining both Fourier phase retrieval and the tomographic problem. We observe an important component of improving the well-posedness of the recovery problem is related to pupil asymmetry. We study the use of algorithms such as a modified Gerchberg-Saxton algorithm and gradient-based methods applied to recovery.
09:30COIMG-143
Rapid wavefront shaping using an optical gradient acquisition, Anat Levin, Technion (Israel) [view abstract]
Wavefront shaping systems enable deep tissue imaging by correcting scattering aberrations, but estimating optimal modulation correction is challenging, since it depends on the unknown tissue structures. Most current methods use slow coordinate descent algorithms, which sequentially scan all modulation parameters and query them independently, thus their complexity scales prohibitively with the number of parameters. We introduce a rapid wavefront shaping system, replacing coordinate descent with gradient descent optimization. To this end, our system acquires a gradient vector, which allows simultaneous update of all modulation parameters. We start with a non-invasive, guide-star-free score function to assess modulation quality and analytically derive its gradient with respect to all modulation parameters. Although the gradient depends on unknown tissue structure, we show it can be inferred from optical measurements. This enables fast, high-resolution wavefront correction with complexity independent of parameter count. We demonstrate the system s effectiveness in correcting aberrations in a coherent confocal microscope.
09:50COIMG-144
Consensus multi-image super resolution, John Mulcahy-Stanislawczyk, Sandia National Laboratories (US); Amber Dagel, Sandia National Laboratories (US); Kelvin Lee, Sandia National Laboratories (US); Suyeon Shin, Sandia National Laboratories (US); Eric Shields, Sandia National Laboratories (US); Yufang Sun, Purdue University (US); Charles Bouman, Purdue University (US); Gregery Buzzard, Purdue University (US) [view abstract]
Wavefront Estimation and Imaging II
11:00 - 12:00
Grand Peninsula C
11:00COIMG-145
Digital adaptive optics for reflective coherent imaging of extended objects, Casey Pellizzari, US Air Force Academy (US); David Strong, Strong Inc.; Tyler Hardy; Christopher Metzler; Mark Spencer [view abstract]
In this work, we introduce and experimentally demonstrate a digital adaptive optics (DAO) framework capable of forming speckle-free high-resolution images of coherently-illuminated diffuse scenes through severe optical aberrations without a guidestar nor interferometric measurements. Computational wavefront sensors (WFss) capture a sequence of phase or amplitude modulated measurements and then recover the complex field of incoming light using a phase retrieval algorithm. This computational approach to wavefront sensing allows these systems to recover higher resolution fields than traditional WFss without relying on complex and sensitive interferometric measurements. However, existing computational WFss are generally restricted to transmission-mode imaging geometries and, due to speckle noise, cannot image coherently-illuminated diffuse reflective targets. In this work we introduce an algorithmic framework that combines speckle-aware physical models, implicit neural signal representations, and denoiser-based image regularization to overcome these limitations.
11:20COIMG-146
How asymmetric should a pupil be? Design trade-offs for single measurement wavefront estimation, Ali Almuallem, Purdue University (US); Nicholas Chimitt, Purdue University; Stanley Chan, Purdue University [view abstract]
Wavefront estimation from a single intensity measurement is a fundamentally ill-posed inverse problem closely related to phase retrieval. Recovering the phase is essential to adaptive optics and is often done using specialized optical hardware such as Shack-Hartmann sensors. Computational approaches are challenged by ambiguous solutions that make the inverse mapping one-to-many, restricting the usage of supervised machine learning methods. Recent results highlights that asymmetric pupils can break these ambiguities and enable single measurement wavefront estimation, but the degree of asymmetry required and its impact on system performance remains unclear. In this work, we empirically study how pupil asymmetry influences phase identifiability and downstream optical correction performance. Through simulation, we characterize a fundamental trade-off: highly asymmetric pupils improve identifiability but make the system diffraction-limited while weakly asymmetric pupils preserve optical resolution but remain aberration-limited. These results provide practical guidance for selecting pupil asymmetry to balance identifiability and corrected system performance in single-measurement wavefront estimation.
11:40COIMG-147
Reflective synthetic aperture via deep image prior, Gordon Hageman, The University of Arizona (US); Andre Van Rynbach, AFRL; David Brady, [view abstract]
This work presents a computational framework for snapshot reflective aperture synthesis using a sparse, static camera array. Conventional synthetic aperture techniques rely on dense scanning for spectral overlap, but snapshot systems inherently produce sparse k-space sampling due to physical camera constraints. This sparsity creates spectral gaps that render traditional phase retrieval ineffective. To address this, we employ the Deep Image Prior (DIP) as an implicit neural regularizer. By modeling the coherent reflection where topology is encoded as phase, we synthesize a numerical aperture significantly larger than any individual sensor. The untrained DIP framework exploits the inductive bias of convolutional networks to bridge unsampled spectral regions, enabling robust phase recovery without labeled datasets. Numerical simulations of an 8-camera system demonstrate the successful recovery of sub-diffraction features under sparse overlap conditions. We further quantify the impact of alignment errors, specifically sub-pixel registration and angular misalignments, to evaluate the framework's stability. The results indicate that the network's structural constraints effectively regularize the ill-posed synthesis problem, offering a robust path for high-resolution snapshot imaging.
Neutron Imaging I
Session Chair: Shimin Tang, Oak Ridge National Laboratory
12:00 - 12:40
Grand Peninsula C
12:00COIMG-148
Diffusion prior regularized implicit neural representation for neutron CT reconstruction of concrete microstructures, Maliha Hossain, Oak Ridge National Laboratory (US); Haley Sullivan, Oak Ridge National Laboratory; Singanallur Venkatakrishnan, Oak Ridge National Laboratory (US); Timofey Efimov, Oak Ridge National Laboratory; Patxi Fernandez-Zelaia, Oak Ridge National Laboratory (US); Amir Ziabari, Oak Ridge National Laboratory (US) [view abstract]
We introduce DiffusiveINR (DINR), a diffusion-driven implicit neural representation framework trained on synthetic data to enable high-quality reconstruction from sparse-view neutron computed tomography (CT). We demonstrate its effectiveness on synthesized concrete microstructures, where conventional reconstruction degrades severely when reducing from 256 views to as few as 4 or 8 views. DINR substantially mitigates artifacts and achieves significant improvements in both PSNR and ssIM, enabling accurate microstructural characterization under extreme data constraints. In future work, we will include a comparative analysis with state-of-the-art sparse-view reconstruction methods and present preliminary results on real neutron CT datasets.
12:20COIMG-149
Bringing virtual reality into the neutron tomography workflow, Leslie Butler, Louisiana State University (US) [view abstract]
Is Virtual Reality a toy or a useful tool in neutron tomography? As of October 2025, this remains an unanswered question. The talk will describe our tour through VR with these points: Purchase and experiences with Apple Vision Pro (version 1) and Meta Quest 3 Virtual Reality headsets CAD: Using Apple Vision Pro to assess CAD drawings of a neutron/gamma ray endstation designed for a terawatt, femtosecond laser system. The Shapr3D CAD software has tight integration with Apple Vision Pro. Medical MRI and CT scans: In the absence of published workflow/software connections for medical DICOM files to the Apple Vision Pro, we developed our own workflow/software. The essential tools are Wolfram (Mathematica), Blender, and Apple Freeform. Siemens Cinematic Reality app for the Apple Vision Pro has recently been reviewed (arXiv:2506.04972v1, July 2025). The app is intended for visualizing MRI and CT scans for the medical community, much as desired for neutron tomography community. The performance of the app in applications as used by surgeons and patients was favorably reviewed by a fourteen subjects. The envisioned use cases selected based on relevance to neutron tomography include: Education and training, Preoperative planning, and Surgical discussions The talk will describe the application of the Apple Vision Pro to ORNL MARS neutron tomography data of a wire arc additive manufactured test object, and inspection of the undesired corrosion contaminants in the test object.
Neutron Imaging II
Session Chair: Shimin Tang, Oak Ridge National Laboratory
15:30 - 17:30
Grand Peninsula C
15:30COIMG-151
COIMG KEYNOTE: Advanced materials characterization via neutron resonance imaging (NRI) and spectroscopy (NRS) at LANSCE, Tsviki Hirsh, Soreq Nuclear Research Center (Israel); Adrian Losko, Forschungs-Neutronenquelle Heinz Maier-Leibnitz (Germany); Alexander Wolfertz, Forschungs-Neutronenquelle Heinz Maier-Leibnitz (Germany); Jean-Christophe Bilheux, Oak Ridge National Laboratory (US); Tim Jaeger, Technical University Darmstadt (Germany); Chen Zhang, Oak Ridge National Laboratory (US); Nicholas Mendez, Los Alamos National Laboratory (US); Sven Vogel, Los Alamos National Laboratory (US) [view abstract]
Neutron Resonance Imaging (NRI) is a non-destructive method that uses isotope-specific variations in neutron cross-sections to map isotopic distributions within materials. Distinct resonance features act as spectral fingerprints, allowing 2D and 3D imaging of isotopic composition without damaging samples. Event-mode neutron imaging has advanced NRI into a quantitative tool by recording individual neutron interactions in time and space. This enables energy-resolved measurements, background discrimination, and improved signal quality, enhancing spectral fitting accuracy. These advances extend NRI into the Neutron Resonance Spectroscopy (NRS) regime, providing spatially resolved insights into temperature and chemical environments through Doppler broadening and resonance line-shape analysis.Neutron Resonance Imaging (NRI) is a non-destructive method that uses isotope-specific variations in neutron cross-sections to map isotopic distributions within materials. Distinct resonance features act as spectral fingerprints, allowing 2D and 3D imaging of isotopic composition without damaging samples. Event-mode neutron imaging has advanced NRI into a quantitative tool by recording individual neutron interactions in time and space.At the Los Alamos Neutron Science Center (LANSCE), optically based event-mode detectors known as LumaCams have been developed to further improve NRI performance. Using Timepix3 sensors coupled to scintillators, LumaCams provide precise spatial and temporal resolution while enabling discrimination between neutron and gamma-ray events. This capability significantly reduces background noise, improves signal quality, and enhances the reliability of energy-resolved transmission spectra, supporting more accurate spectral fitting and isotope quantification.These advances extend NRI into the Neutron Resonance Spectroscopy (NRS) regime, providing spatially resolved insights into temperature and chemical environments through Doppler broadening and resonance line-shape analysis. Together with ongoing developments in analysis tools and multi-modal integration, NRI and NRS now offer a robust framework for non-destructive, isotope-specific characterization of complex materials under both ambient and extreme conditions.
15:50COIMG-152
Computational methods for neutron resonance imaging with sparse spectral data: R-matrix forward models and regularized reconstruction, Chen Zhang, Oak Ridge National Laboratory (US); Luiz Leal, Oak Ridge National Laboratory (US); Jean-Christophe Bilheux, Oak Ridge National Laboratory (US); Alexander Long, Los Alamos National Laboratory (US); Shimin Tang, Oak Ridge National Laboratory (US); Hassina Bilheux, Oak Ridge National Laboratory (US) [view abstract]
Neutron resonance imaging exploits isotope-specific resonance signatures in time-of-flight transmission measurements to produce spatially-resolved isotopic maps. At pulsed neutron sources, 2D detectors acquire transmission spectra at each pixel, creating hyperspectral imaging data with spatial and energy dimensions. However, limited per-pixel neutron flux in spatially-resolved measurements results in sparse neutron counts, posing significant challenges for pixel-by-pixel spectral analysis. This work presents a joint spatial-spectral reconstruction framework addressing the inverse problem of reconstructing spatially-resolved isotope distributions from under-sampled resonance imaging data. The approach combines physics-based R-matrix forward models with Bayesian inference, validated through integrated spectrum analysis at the VENUS beamline. Joint spatial-spectral reconstruction strategies under development including non-negative matrix factorization on local pixel neighborhoods and hierarchical subdivision with adaptive spatial resolution couple neighboring pixels through spatial regularization and physical constraints to enable 2D and 3D isotopic imaging for nuclear materials characterization and safeguards applications.
16:10COIMG-153
Simultaneous X-ray and neutron radiography using event-mode imaging techniques, Nicholas Mendez, Los Alamos National Laboratory (US); Alexander Long, Los Alamos National Laboratory (US); Patrick Feng, Sandia National Laboratory (US); Tsviki Hirsh, Soreq Nuclear Research Center (Israel); Adrian Losko, Forschungs-Neutronenquelle Heinz Maier-Leibnitz (Germany); Sven Vogel, Los Alamos National Laboratory (US); Charles Leak, Nevada National Security Site (US) [view abstract]
Event-mode imaging enables real-time recording of individual scintillation photons, allowing the temporal response of a scintillator to be fully captured and analyzed. This capability makes it possible to leverage pulse-shape differences for neutron and gamma discrimination within a single detector system. We present recent results using Timepix-based event-mode cameras coupled with several scintillator materials, including organic glass scintillator (OGS), CLYC, and CLLBC, under mixed-field exposure from portable neutron and gamma sources. The event-driven data structure enables detailed analysis of photon arrival statistics and spatial clustering, facilitating particle-type identification and simultaneous dual-modality radiography. Ongoing measurements are focused on quantifying differences in light yield, decay behavior, and event topology across these scintillators to assess their suitability for event-mode operation. These studies will inform the development of compact, high-resolution systems capable of real-time, dual-modality imaging with portable sources.
16:30COIMG-154
Computational imaging challenges and AI solutions for hyperspectral neutron imaging at VENUS instrument, Shimin Tang, Oak Ridge National Laboratory (US); Kevin Yahne, Oak Ridge National Laboratory (US); Greg Guyotte, ; Ray Gregory, ; Jean-Christophe Bilheux, ; Hassina Bilheux, [view abstract]
The Versatile Neutron Imaging Instrument (VENUS) at the Spallation Neutron Source is a neutron imaging beamline designed for advanced two- and three-dimensional imaging in materials science, energy research, nuclear materials, and biology. Following a series of commissioning activities, VENUS began routine user operations in 2025, supporting applications such as non-destructive testing of metals and ceramics, strain mapping, phase identification, and isotopic imaging. A key capability of VENUS is hyperspectral time-of-flight (ToF) neutron imaging. This technique provides spatially resolved spectral information with thousands of wavelength or energy channels per dataset. While this enables detailed material characterization, it also leads to long acquisition times and very large, high-dimensional datasets. Conventional reconstruction and image processing methods are often insufficient for handling these data efficiently. To address these challenges, we developed an AI-driven autonomous hyperspectral neutron computed tomography framework, termed HyperCT. HyperCT integrates hyperspectral CT reconstruction, image quality evaluation, and adaptive selection of projection angles into a closed-loop workflow. Based on intermediate reconstruction results, the system determines subsequent measurement angles that improve information content and reduce redundant data acquisition. In parallel, VENUS serves as a development platform for advanced AI methods in neutron imaging. Current efforts include deep-learning-based fine segmentation with limited training data, analysis of high-dimensional neutron spectral information, and physics-informed data interpretation. These developments support efficient use of beam time and enable new experimental workflows. This presentation introduces the VENUS instrument and the HyperCT framework and discusses opportunities for applying computational imaging and artificial intelligence methods to large-scale neutron imaging experiments.
16:50COIMG-155
Dehydration-rehydration: A training-free machine learning algorithm for reliable hyperspectral denoising, Mohammad Samin Nur Chowdhury, ; Diyu Yang, (US); Shimin Tang, (US); Singanallur Venkatakrishnan, (US); Hassina Bilheux, (US); Gregery Buzzard, (US); Charles Bouman, [view abstract]
Hyperspectral neutron imaging is a powerful non-destructive technique for analyzing material spectral characteristics by resolving time-of-flight transmission data into thousands of wavelength-specific bands. However, the technique faces a critical limitation: each band has an extremely low signal-to-noise ratio (SNR), even after long acquisition times. Consequently, per-pixel spectral profiles are often overwhelmed by noise. Traditional denoising methods oversmooth spatial or spectral features, while standard machine learning approaches are impractical due to the lack of training data.We propose Dehydration Rehydration, a training-free machine learning algorithm for fast, large-scale hyperspectral denoising that preserves both spatial and spectral details. Dehydration performs dimensionality reduction using unsupervised non-negative matrix factorization (NMF), which projects the noisy hyperspectral data into a compact, low-dimensional subspace. Because noise patterns are uncorrelated and non-repetitive across pixels, they cannot be efficiently represented in such a compact space and are naturally rejected, while the dominant spectral structures shared across pixels are retained. Rehydration then projects the subspace data back into the full hyperspectral domain.Experiments on simulated and measured datasets demonstrate over 30 dB SNR improvement without compromising resolution. By enabling high-quality data reproduction from significantly lower neutron counts, this method can substantially reduce acquisition time and beamline resource usage.
Computational Imaging Panel: AI for Science: Approaching the Singularity
18:00 - 19:05
Grand Peninsula C
Panel: AI for Science: Approaching the Singularity, [view abstract]
Moderator:Greg Buzzard, Purdue University. AlphaFold's solution of the protein folding problem demonstrated that AI can dramatically accelerate the advancement of fundamental science. But a wide array of questions remain. How should we adapt AI research to science, and how should we adapt scientific investigation to AI methods? This panel attempts to address the foundational question of 21st century science and provide some useful insights into how we should proceed as a community.
THURSDAY 5 MARCH 2026
Multimedia
Session Chair: Maggie Zhu, Purdue University
08:30 - 10:30
Grand Peninsula C
08:30COIMG-156
CompressAI and CompressAI-Vision: Open-source software to evaluate compression methods for humans and machines, Hyomin Choi; Mateen Ulhaq; Fabien Racape, [view abstract]
With the rapid rise of neural network (NN) based computer vision applications that rely on image and video data, there is a growing demand for compression technologies optimized for both humans and machines. CompressAI and CompressAI-Vision are two open-source libraries developed to address this emerging need and to support a wide range of research problems. CompressAI provides a flexible PyTorch-based framework for developing, training, and evaluating learned image and video compression models, facilitating research and experimentation in neural compression. Building on this foundation, CompressAI-Vision extends these capabilities to the domain of coding for machines, offering a standardized platform to evaluate compression performance in terms of bit-rate versus task accuracy across different vision tasks and datasets. The platform supports both remote and split inference scenarios, enabling comprehensive benchmarking of compression models for downstream vision applications. Together, CompressAI and CompressAI-Vision provide a unified environment that bridges learned compression research and practical standardization for both human and machine vision systems. The software is openly available at https://github.com/InterDigitalInc/CompressAI and https://github.com/InterDigitalInc/CompressAI-Vision.
08:50COIMG-157
AI-driven video transformation: Compression, understanding, and streaming, Zoe Liu, Visionular Inc (US) [view abstract]
Video applications, from VOD to low-latency streaming, are expanding at an unprecedented pace. This rapid growth is creating urgent demand for intelligent, efficient, and scalable video processing technologies and solutions. In this talk, I will introduce recent advances in AI-driven video transformation across compression, enhancement, quality assessment, content understanding, and streaming.We will highlight how modern video codecs are increasingly shaped by machine learning, with a focus on emerging standards such as the AOM AV1 and AV2 projects. Neural tools are improving coding efficiency, yet they also present practical challenges in hardware feasibility, complexity management, and standardization timelines. I will contrast these developments with progress in real-world deployment and practical implementation.To ground these trends, we will present the applied work from our Visionular team. Our AI-driven video solutions provide standard-compliant compression, adaptive reframing, content-aware analysis, and streaming optimization for customers including Snap, Pinterest, NBCUniversal, HUDL, ABS-CBN, and ShareChat. We will also address the emerging challenge of compressing GenAI-generated content, which demands perceptually adaptive strategies beyond traditional encoders. In addition, I will share our GPU-accelerated transcoding pipeline, built on Nvidia s NVENC module, for both VOD and live workflows. This approach integrates no-reference VQA, content pre-analysis and enhancement, and ROI-based rate control and configuration.Overall, this talk illustrates how AI is reshaping the entire video pipeline, including compression, understanding, enhancement, and delivery, and how research innovation is transitioning into industrial-scale impact.
09:10COIMG-158
MPEG AI technologies and standards, Shan Liu, Tencent (US) [view abstract]
09:30COIMG-159
On the development of open-source video codecs AV1 and AV2, Yunqing Wang, Google (US); Urvang Joshi, Google (US); Mohammed Sarwer, Google (US); Jianle Chen, Google (US); Lester Lu, Google (US); Bohan Li, Google (US); Stan Vitvitskyy, Google (US); In-Suk Chong, Google (US); Jingning Han, Google (US); Debargha Mukherjee, Google (US) [view abstract]
With the recent explosion in online video consumption, up to 75% of all Internet bandwidth (according to some estimates) is used by video. Consequently, improving bandwidth efficiency through advances in video compression technology has become quite important. The Alliance for Open Media (AOM) industry consortium released its first open video codec AV1 in 2018, and a more advanced second codec AV2 in 2025. Both codecs have been developed in the open-source software model.Modern video codecs are extremely complex with many inter-connected modes, and complex interactions between them. The encoder software is expected to perform a very complex optimization to determine a near optimal collection of coding decisions and a corresponding compressed bitstream, providing the best quality of encoding at the least possible bitrate and computational cost. Generally speaking, this is a very complex rate-distortion-complexity optimization problem in a discrete space, therefore making it an integer programming problem that is completely intractable to solve globally. Any practical solution must use many local approximations and intelligent guesses to prune out possible paths and options. To this end, many ML based algorithms using contextual and image/video features are increasingly used in modern codecs.The decoder side of the codec, on the other hand, must be optimized to reconstruct video at blazing fast video throughputs: for example 4K resolution at 60 frames per second. To make this possible, sophisticated SIMD instructions as well as parallelism are extensively used, which makes the software very complex.Typically, the software development for a codec has two distinct phases - first while the codec is under undevelopment, and second, when the syntax of the codec is finalized. The first phase is necessary to even be able to run simulations at a reasonable time. Besides, in the first phase we also would want to tune the actual coding tools and algorithms to make it even possible to optimize better on the encoder and decoder sides. The second phase is where optimizations in earnest can happen as the new codec is productized. This talk will provide an overview of the software development process for the recent video codecs AV1 and AV2, that largely followed this model.
9:50COIMG-Panel3
Panel: Software Design and Implementation in Multimedia Technologies, Systems, and Applications, Maggie Zhu, Purdue University (US) [view abstract]
Multimedia technologies demand software that can manage diverse data types, real-time performance, and user-centered design. In this panel discussion, we will discuss how thoughtful design enables robust, innovative solutions in this rapid advancing field and the challenges and opportunities of AI reshaping the future of video codecs. Moderator: Maggie Zhu. Panelists: Hyomin Choi, InterDigital; Zoe Liu, Visionular; Shan Liu, Tencent; Yunqing Wang, Google.
Imaging Pipelines II (Joint Session with Machine Learning for Scientific Imaging)
11:00 - 13:00
Grand Peninsula C
11:00COIMG-160
3D field of junctions: A noise-robust, training-free structural prior for volumetric inverse problems, Sara Fridovich-Keil (US) [view abstract]
Volume denoising is a foundational problem in computational imaging, as many 3D imaging inverse problems face high levels of measurement noise. We propose a fully volumetric version of Field of Junctions, an explicit structural prior that has shown strong denoising properties for 2D images. Like its 2D counterpart, our 3D Field of Junctions (3D FoJ) extracts and preserves sharp boundary structure by optimizing a junction of 3D wedges that best explain each 3D patch of a full volume, while encouraging consistency between overlapping patches. As a denoising prior, 3D FoJ (i) requires no training data, and thus precludes the risk of hallucination, (ii) preserves and enhances sharp edge and corner structures in 3D, even under low signal to noise ratio (SNR), and (iii) can be used as a drop-in denoising representation via projected gradient descent for any volumetric inverse problem with low SNR. We demonstrate successful volume denoising with 3D FoJ across three diverse 3D imaging tasks: low-dose X-ray computed tomography (CT), denoising volumes from cryogenic electron tomography (cryo-ET), and denoising point clouds such as those from lidar with miscalibration or adverse weather. Across these domains, 3D FoJ outperforms a mixture of classical and neural denoisers.
11:20COIMG-161
Near real-time stitching for a heterogenous camera array, Adel Al-Ghazwi, University of Arizona, Wyant College of Optical Sciences (US); Gordon Hageman, University of Arizona, Wyant College of Optical Sciences (US); Kimberly Doty, University of Arizona, Wyant College of Optical Sciences (US); David Brady, University of Arizona, Wyant College of Optical Sciences (US) [view abstract]
We demonstrate a streamlined pipeline for near real-time image stitching from a heterogeneous camera array. Methods using static homography struggle with dynamic scenes and parallax-induced artifacts. Our approach overcomes these limitations by integrating modern lightweight neural networks, ALIKE and LightGlue, for fast robust feature matching. These models are converted into an optimized TensorRT engine, allowing for periodic homography re-estimation during the capture stream. We also address parallax errors by replacing a single global transformation with local mesh-based homographies, which provide more precise, non-rigid alignment. This pipeline delivers a dynamic and accurate stitching pipeline with low latency that is adaptable to different camera array systems.
11:40COIMG-162
Trace2SCAD: Converting images Into parametric OpenSCAD models, Henry Dietz, University of Kentucky (US) [view abstract]
Ironically, it is often useful to manipulate 2D graphics for 3D printing. Trace2SCAD was originally written to convert 2D diagrams into printed structures that could be understood by a blind student feeling the surface. However, by its initial open source release in January 2015, it had already been applied to convert pixel-mapped images into parametric 3D models and 3D-printable image renderings (e.g., lithophanes). The original version was a shell script using a variety of helpers, especially potrace. The new version is pure C++ code using the OpenCV library and incorporating many new features, including the ability to render color images as 3D models to be printed for viewing by reflected light (similar to the filament paintings produced by HueForge). It is also significant that the output generated is not mere 3D shapes, but parametric 3D models coded in the OpenSCAD programming language. This paper describes the functionality and algorithms used in the new Trace2SCAD.
12:00COIMG-163
NUTIK: A testbed for functional post-capture manipulation of time and gain, Paul Eberhart, University of Kentucky (US); Henry Dietz, University of Kentucky (US) [view abstract]
Conventional photographic imaging technologies fix the time, interval, and gain of an exposure at the instant of capture, spoiling images if these parameters are not correctly anticipated. Techniques, such as TDCI (Time-Domain Continuous Imaging) allow the integration of an image to be decoupled from the capture of scene data, creating an opportunity to not only adjust the timing and gain of an image after the fact, but manipulate those parameters in previously impossible ways.To facilitate exploration of these new dimensional freedoms, we have created NUTIK, a tool which allows scene data to be captured and computationally post-processed to expose images with user control over the time interval being sampled and the gain of integration, not just for each image rendered but for every site in each rendered image. This paper documents the design and operation of NUTIK, and makes an initial exploration of useful and interesting new photographic techniques enabled by such a tool.
12:20COIMG-164
Arbitrary spatial sampling with the B-FFT for efficient exoplanet imaging simulations, Jamila Taaki, University of Michigan (US); Farzad Kamalabadi, Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign (US); Athol Kemball, Department of Astronomy, University of Illinois at Urbana-Champaign (US); Lia Corrales, Department of Astronomy, University of Michigan (US); Alfred Hero III, Department of Electrical Engineering and Computer Science, University of Michigan (US) [view abstract]
Direct imaging of Earth-like exoplanets in the habitable zone of a Sun-like star requires extreme starlight suppression, typically on the order of 10-8 to 10-10. Starshades are petal-shaped external occulters, some 10 s of meters across, flown in formation with space telescopes to suppress on-axis starlight while preserving off-axis planetary signals. Accurate computational simulations of starshade Fresnel diffraction are essential for mission design studies and quantifying performance metrics for future mission concepts such as NASA's Habitable Worlds Observatory (HWO). However, these simulations demand moving between fine spatial resolutions over multiple planes of propagation and across broad wavelength bands. Traditional fast Fourier transforms (FFTs) require large zero-padding factors to accomplish this, increasing computational cost. In this work, we present a general optical model for starshades and demonstrate the adaptation of the Bluestein Fast Fourier Transform (B-FFT) to enable efficient, high-fidelity imaging simulations of starshades at arbitrary spatial scales without zero-padding, thereby achieving greater efficiency than the FFTs or discrete Fourier transforms (DFTs).
12:40COIMG-165
Parallelizable iterative CT reconstruction, Sri Ragha Sai Sowmya Seeram, Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana (US); Charles Bouman, Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana (US); Gregery Buzzard, Department of Mathematics, Purdue University, West Lafayette, Indiana (US) [view abstract]
Volumetric X-Ray Imaging
Session Chair: Greg Buzzard, Purdue University
14:00 - 16:00
Grand Peninsula C
14:00COIMG-167
Advances in deep learning-based image reconstruction, resolution recovery, and artefact removal for X-ray microscopy, Matthew Andrew, Carl Zeiss; Faguo Yang, ; Moran Xu, Carl Zeiss (US); Shiqi Xu, Carl Zeiss (US); Tianqi Zhang, Carl Zeiss (US); Zeyu Zhou, Carl Zeiss (US); Andriy Andreyev, Carl Zeiss [view abstract]
Recent advances in deep learning have revolutionized X-ray microscopy, enabling substantial improvements in image reconstruction, resolution recovery, and artefact/noise removal. This paper synthesizes key findings and methodologies from recent studies, presenting a comprehensive overview of novel workflows including deep convolutional neural networks for resolution recovery, fully automated training schemes, and synthetic prior image restoration. Benchmarking across a wide spectrum of materials and sample types demonstrates significant gains in image quality, quantitative accuracy, and throughput, overcoming longstanding limitations of conventional techniques. The implications for generalizability, automation, and future research directions are discussed.
14:20COIMG-168
Fast large-scale model-based iterative tomography via exploiting mathematical structure, hierarchical optimization, smart initialization, and distributed GPU computing, Dinesh Kumar, Lawrence Berkeley National Labortatory (US); Jeffrey Donatelli, Lawrence Berkeley National Laboratory (US) [view abstract]
Although MBIR yields high-quality reconstructions through prior-informed optimization, its computational cost has traditionally limited its wider adoption. We address this limitation by expressing the Radon transform and its adjoint via non-uniform fast Fourier transforms to reduce computational complexity relative to conventional projection-based methods. We further accelerate computation by employing a multi-GPU system for parallel processing.In this work, we introduce a multi-strategy, further acceleration our Fourier domain framework, comprising of four major strategies: (1) a reformulation of the MBIR forward and adjoint operators that exploits their Toeplitz structure for efficient Fourier-domain computation; (2) a smarter initialization strategy that uses back-projected data filtered with a simple ramp filter as the starting guess; (3) a hierarchical multi-resolution reconstruction approach that first solves the problem on coarse grids and progressively transitions to finer grids using Lanczos interpolation; and (4) a distributed-memory implementation using MPI that enables scaling to large HPC systems. style="font-family: "Calibri", sans-serif; font-size: 11pt; line-height: 115%">Together, these innovations significantly reduce iteration counts, improve parallel efficiency, and make high-quality MBIR reconstruction practical for large-scale tomographic imaging. These advances open the door to near-real-time imaging with applications ranging from in-situ, in-operando and time-dependent experiments.
14:40COIMG-169
Science at the speed of light: New computational infrastructure for performing real-time analyses on tomographic imaging data at the Advanced Light Source, Elizabeth Clark, Lawrence Berkeley National Laboratory (US); Pradyumna Elavarthi, Lawrence Berkeley National Laboratory, University of Cincinnati; Alex Hexemer, Lawrence Berkeley National Laboratory; Tanny Chavez, Lawrence Berkeley National Laboratory; Wiebke Koepp, Lawrence Berkeley National Laboratory; Petrus Zwart, Lawrence Berkeley National Laboratory; David Abramov, Lawrence Berkeley National Laboratory; Dylan McReynolds, Lawrence Berkeley National Laboratory; Sam Welborn, Lawrence Berkeley National Laboratory; Lee Yang, Lawrence Berkeley National Laboratory; Damon English, Lawrence Berkeley National Laboratory; Kuldeep Chawla, Lawrence Berkeley National Laboratory; Dilworth Parkinson, Lawrence Berkeley National Laboratory (US)) [view abstract]
Driven by the team science mentality at Berkeley Lab and the greater DOE light source ecosystem, we have harnessed the power of a multi-institutional interdisciplinary collaborative force to systematically target sticking points in the most urgent data bottlenecks for time-resolved imaging at the Advanced Light Source s micro-tomography beamline 8.3.2. We have addressed issues at every stage of the pipeline: image writing, transfer, visualization, storage, processing, and analysis. As a result, scientists can now perform data analysis and deliver results from tomographic imaging at the Advanced Light Source nearly in real-time, reducing the time to discovery. Here, we will give an overview of our progress, with a focus on our recent efforts to leverage AI/ML approaches and high performance computing to automate image segmentation for large 3D data sets. Image segmentation is an essential step for most scientific studies at our beamline including analysis, quantification, and the creation of digital twins. The annotated regions of interest from this process can be used for a wide variety of downstream use cases, from computational modeling to virtual reality experiences. This serves as the model for infrastructure development for implementation across synchrotron facilities with high-throughput tomographic imaging capabilities. , [view abstract]
Driven by the team science mentality at Berkeley Lab and the greater DOE light source ecosystem, we have harnessed the power of a multi-institutional interdisciplinary collaborative force to systematically target sticking points in the most urgent data bottlenecks for time-resolved imaging at the Advanced Light Source s micro-tomography beamline 8.3.2. We have addressed issues at every stage of the pipeline: image writing, transfer, visualization, storage, processing, and analysis. As a result, scientists can now perform data analysis and deliver results from tomographic imaging at the Advanced Light Source nearly in real-time, reducing the time to discovery. Here, we will give an overview of our progress, with a focus on our recent efforts to leverage AI/ML approaches and high performance computing to automate image segmentation for large 3D data sets. Image segmentation is an essential step for most scientific studies at our beamline including analysis, quantification, and the creation of digital twins. The annotated regions of interest from this process can be used for a wide variety of downstream use cases, from computational modeling to virtual reality experiences. This serves as the model for infrastructure development for implementation across synchrotron facilities with high-throughput tomographic imaging capabilities.
15:00COIMG-170
Accurate reconstruction of plastic metal objects in X-ray CT imaging, Ziyun Li, (US); Mingqi Yang ; Vijay Sharma ; John Toaquiza; Javad Eshraghi; Gregery Buzzard; Charles Bouman, Purdue University (US) [view abstract]
X-ray CT imaging of objects that contain a combination of metal and plastic materials is challenging because the metal artifacts tend to distort the plastic reconstruction.We present an Iterative Plastic Metal Reconstruction (IPMR) algorithm that reduces artifacts by segmenting the object, estimating the material beam hardening parameters for each material, and then reconstructing the corrected sinogram.Results on scanned data indicate the IPMR algorithm can significantly improve the quality of plastic reconstruction while maintaining the quality of the metal.
15:20COIMG-133
Multimodal volumetric imaging with a high-resolution plenoptic hard X-ray laboratory system, Shiqi Xu, Zeiss (US) [view abstract]
We report a lab-based high-resolution tomography system for simultaneous retrieval of absorption, phase, and darkfield signals in the hard X-ray spectrum. The system is empowered by a high-brilliance micro-focal source, alone with a small-pixel microscope objective-based detector system. We demonstrate complementary modalities for imaging a variety of samples in botany, entomology, and materials science.
15:40COIMG-174
Characterization of temporal dynamics of large aperture and deformable mirrors, Vladimir Markov, AS&T (US) [view abstract]
Large Aperture Mirrors (LAM) of various designs are essential for imaging systems, ground-based and space telescopes, as well as for laser beam director and control modules. The need exists in characterization of multiple modality dynamics of recently developed membrane and 3D printed mirror structures. Knowledge of this dynamic is essential for defining LAM s optimal configuration, balancing and eventually for predicting the operational efficacy of future systems. This presentation discusses the Whole-Field Laser Doppler Vibrometer (WF-LDV) - the newly developed instrument capable of instant characterization of LAM s vibrational spectra and full dynamic of its membrane-like mirror surface. WF-LDV allows a high accuracy mapping of the temporal dynamics of the mirror surface translation in the multi-kHz frequency band. Innovative WF-LDV design supports measurements of multiple modalities of optical metrology tailored specifically for LAM design.