450 papers
This section explores the diverse scientific discoveries emerging from the alphabet soup of research, covering everything from computational chemistry to gravitational waves. These studies span a wide range of disciplines, bridging gaps between theoretical models and real-world applications in physics, engineering, and the life sciences. By gathering these works, we aim to show how fundamental questions are being answered across this broad spectrum of inquiry.
Every new preprint in this category originates directly from arXiv, the world's leading open-access repository. At Gist.Science, we process each of these fresh submissions as they arrive, transforming dense academic manuscripts into accessible content. You will find both plain-language explanations for curious readers and detailed technical summaries for specialists, ensuring that the latest breakthroughs are understandable to everyone regardless of their background.
Below are the latest papers and summaries in this collection, updated daily as new research becomes available.
This paper introduces Instant NuRec, a feed-forward neural model that rapidly reconstructs fully simulatable 3D Gaussian Splatting driving scenes from multi-view logs in a single forward pass, significantly outperforming existing methods in both speed and rendering quality while supporting dynamic elements and non-pinhole cameras.
This paper proposes an online neural space-time memory framework for dynamic novel view synthesis that achieves real-time performance by decoupling periodic memory updates from per-frame application, thereby balancing persistent long-horizon reconstruction with strict computational constraints.
This paper proposes a neural inverse rendering framework that jointly optimizes neural fields for optical properties and the full light field to recover spatially varying scattering, absorption, and phase functions from multi-view images by enforcing global illumination through a residual objective derived from the Radiative Transfer Equation.
This paper introduces Adaptive Fluid Cohomology, a framework that integrates dynamic spatial and temporal refinement with a robust harmonic basis transfer method to efficiently and stably simulate inviscid, incompressible flows on curved surfaces, achieving significant memory savings while maintaining accuracy even on poor-quality meshes.
RegHead introduces a fast, feed-forward framework that constructs interpretable, semantic blendshape sets for animatable non-humanoid head avatars by leveraging a large-scale dataset and a dense stochastic anchor motion representation to achieve high-fidelity, real-time expression retargeting from human tracking signals.
AcadGIS is a free, open-source Python package that streamlines the creation of reproducible, publication-ready academic maps by consolidating fragmented GIS workflows into a single, high-level import that handles boundary access, cartographic styling, and export without requiring desktop GIS expertise.
This paper proposes a lightweight method for Implicit Neural Representations (INRs) that reformulates regression as a classification task by discretizing targets into bins, enabling efficient and flexible modeling of complex error distributions for improved reconstruction quality and uncertainty estimation without relying on expensive computations or rigid parametric assumptions.
The paper introduces Spatula, a system that generates on-demand, in-situ interfaces for fine-grained attribute control in motion graphics, demonstrating through a user study and cross-domain applications that this approach effectively supports iterative creative refinement.
The paper introduces 3D-DefectBench, a comprehensive benchmark and factorial study demonstrating that the reliability of automated 3D defect detection depends on the entire evaluation pipeline—including camera protocols and prompt schemas—rather than just the underlying vision-language model, while identifying a cost-effective six-view RGB setup and highlighting the performance gap between current AI judges and human labelers.
The paper presents MAGIC, a four-stage prompt-to-project system that leverages large language models to automatically generate navigable, consistent multi-scene game worlds with functional transitions, addressing key challenges in cross-scene consistency, in-scene navigability, and transition validation through a novel pipeline and evaluation agent.