About aiXiv
What is aiXiv?
aiXiv is an open-access archive and distribution service for scientific research papers. Our platform provides free, unrestricted access to scholarly articles with a particular focus on computational imaging, physics world models, and related disciplines at the intersection of physics and computer science.
Physics World Model (PWM) -- Toward Imaging System Autonomy
PWM is the evaluation harness and current best methods for computational imaging -- an open, reproducible toolkit that aims to make any imaging system self-specifying, self-diagnosing, and self-correcting.
The computational imaging community has built increasingly powerful reconstruction algorithms, yet deployed systems routinely fail because the assumed forward model diverges from real physics. PWM provides a universal diagnostic and correction framework grounded in the Triad Law: every imaging failure decomposes into exactly three root causes -- recoverability loss, carrier-noise budget violation, and operator mismatch.
Three Pillars
Standardized Infrastructure
The Rail provides the standardized evaluation harness for computational imaging: OperatorGraph IR spanning 64 modalities across 5 physical carriers with 89 validated templates, a 4-scenario evaluation protocol, the Leaderboard for Imaging Physics (LIP-Arena), and a Red Team adversarial verification module.
Universal Diagnostic Theory
The PWM Flagship proves two foundational results: the Finite Primitive Basis Theorem (every imaging forward model decomposes into exactly 10 canonical primitives) and the Triad Decomposition (every failure has three root causes). Autonomous correction recovers +0.8 to +10.7 dB across seven modalities spanning three carrier families. Hardware validation confirms mismatch dominance.
Read the Finite Primitive Basis Theorem paper →
CASP for Computational Imaging
InverseNet is the first cross-modality benchmark for operator mismatch -- inspired by CASP for protein structure prediction (AlphaFold). It evaluates 12 reconstruction methods under a four-scenario protocol across 27 simulated scenes and 9 real hardware captures, totalling over 360 experiments, aiming to catalyze an "AlphaFold moment" for computational imaging.
Applications in Medical & Nuclear Physics
PWM's carrier-based framework naturally extends beyond optical imaging to diagnostic and nuclear medical physics. We are actively developing clinical applications, including two CT quality-control papers:
- CT QC Copilot -- Automated decision-support system reducing per-scanner QC time by 94%
- CT QC Platform -- Open, reproducible framework with versioned CasePacks, extensible to PET/CT and SPECT/CT
The five physical carriers in PWM cover the full spectrum of clinical imaging modalities:
PWM provides the same Triad Law diagnostic framework to any physics-based imaging system: the operator mismatch that causes a CASSI system to lose 16 dB is the same class of failure that causes CT reconstruction artifacts or MRI aliasing -- and the same correction framework applies.
Open Collaboration
PWM is an open-source initiative. We welcome collaborators across all imaging domains to:
- Validate PWM on new imaging modalities (medical, industrial, scientific)
- Add modality-specific operator templates to the OperatorGraph registry
- Contribute reconstruction methods to the InverseNet benchmark
- Co-develop clinical imaging applications
All code, data, and reproducibility manifests are available on GitHub.
Focus Areas
- Computational Imaging (physics.comp-img) -- Forward models, reconstruction algorithms, operator calibration, and imaging system design across all physical modalities.
- Physics of Optics (physics.optics) -- Optical systems, spectral imaging, coded aperture methods, and diffractive optics.
- Medical Physics (physics.med-ph) -- Diagnostic imaging, nuclear medicine, radiation therapy, and clinical image reconstruction.
- Computer Vision (cs.CV) -- Image reconstruction, inverse problems, and visual computing.
- Artificial Intelligence (cs.AI) -- Machine learning methods applied to physical systems and scientific discovery.
Paper Identification System
Every paper on aiXiv is assigned a unique identifier following the format aiXiv:YYMM.NNN, where:
- YY -- Two-digit year of submission
- MM -- Two-digit month of submission
- NNN -- Sequential paper number within that month
For example, aiXiv:2502.001 refers to the first paper submitted in February 2025. This system ensures every paper has a permanent, citable reference.
Operated By
aiXiv is operated by NextGen PlatformAI C Corp, dedicated to building infrastructure for open science and advancing the standardization of computational imaging research.
Contact: integrityyang@gmail.com
← Back to all papers