Specialty
A systematic review finds that most deep-learning brain MRI reconstruction studies did not compare image-fidelity scores with radiologist assessments on the same data, leaving key safety questions unanswered.
Scientific publication4 min read
CF-YOLO, a YOLOv11-based detector built to refine context and fine detail, scored higher than YOLOv11n on CTDD but delivered mixed results on external NEU-DET data.
Scientific publication4 min read
A tiny two-recording test found that a parameter-free color extrapolator beat copying the last frame and every trained comparator on held-out copper in both directions. The air-to-chamber advantage was individually separated from zero, while the chamber-to-air interval included zero.
Scientific publication5 min read
A video AI method called Token-Budget Distillation retained strong benchmark scores after aggressive visual-token compression, although its training still depended on the uncompressed teacher model.
Scientific publication5 min read
An anatomy-aware system called CheXtriev reported stronger case-retrieval scores than global and local comparison methods on selected chest radiographs. The gains were especially notable for several lower-prevalence findings.
Scientific publication4 min read
EfficientNetB0 correctly classified 97.36% of 303 held-out mango images, with eight errors, in a study that also deployed the model through a public web app. The result is an initial within-dataset benchmark, not evidence that the tool will deliver the same performance across regions or improve agricultural decisions.
Journal of Bangladesh Academy of Sciences, vol. 50, Supplement 1, p. 114, 20265 min read
The fixed-round PTD model reported sharp speed gains and higher scores across several video benchmarks, while leaving disjoint events and multiple matching targets largely unexplored.
Scientific publication4 min read
A preprint describes one OCR system trained across 13 Indic scripts; its reported overall character error rate was 6.9%, compared with 8.6% for monolingual models.
Scientific publication4 min read
Tests across nine systems found a sharp drop in numeric precision as prompts asked for more objects. Layout, composition and appearance also mattered, but high-count results were harder to validate.
Scientific publication5 min read
GeoFF3D, a feed-forward system for large UAV image collections, reported higher reconstruction scores than Pi3X + SLRF across nine aerial mapping blocks.
Scientific publication4 min read
A controlled test suggests that visual evidence does not have one fixed ranking: the next useful clue can depend on what has already been acquired and on the target being sought.
Scientific publication4 min read
A computer-vision study reports that a sensor-agnostic training approach improved dense image matching under spectral mismatch, but its real medical demonstrations were qualitative.
Scientific publication4 min read
A two-stage system that denoises partial plant scans before completing missing structure showed lower real-data reconstruction error, while synthetic rankings varied by metric and dataset.
Scientific publication5 min read
A new methods preprint reports that a unified AI model could recover muscle, geometry and movement information from paired 3D tongue meshes, while shuffled inputs sharply reduced performance. The findings are confined to one simulator, anatomy and mesh topology.
Scientific publication4 min read
A preprint reports that LUCAID, an AI system for lung-cancer pathology, showed strong agreement with expert annotations and a panel reference while leaving its effect on routine care unresolved.
Scientific publication5 min read
A two-stage computer-vision pipeline produces a mesh with editable surface maps from multi-view images and tests how well the result holds up under new views and lighting.
Scientific publication5 min read
A preprint reports that SMART performed strongly on benchmark tests combining continuous sign-language recognition with sign spotting, while leaving its real-world communication value untested.
Scientific publication4 min read
A longitudinal MRI benchmark found pair-specific adaptation improved learning-based methods, while optimization-based methods led overall, with too much case-to-case variation for hands-off dose accumulation.
Scientific publication4 min read
A lightweight method called DEX improved benchmark results on indoor and outdoor fisheye images while leaving the original vision backbones frozen.
Scientific publication4 min read
The paper reports that its dual-feedback version ranked highest in benchmark scoring, while its evidence stops short of testing customer or business outcomes.
Scientific publication4 min read
A proposed computer-vision framework scored better than the strongest listed baseline at linking the same person across paired ground and aerial images, but the benchmark remains difficult and broader generalization is untested.
Scientific publication4 min read
ZipMVS aims to reduce the memory burden of multi-view stereo by using fewer, pixel-adaptive depth hypotheses, and reports competitive benchmark quality with mixed speed and memory results.
Scientific publication3 min read
An arXiv preprint reports that Ex-Sim(3)-Reg produced higher inlier-ratio and registration-recall scores than a baseline across disturbed, cross-dataset and low-inlier indoor tests. Its authors also report a speed trade-off as scale settings change and limit the theoretical recovery claim to a candidate subset of true inliers.
Scientific publication4 min read
A source-fully-free adaptation method combines caption guidance with class anchors, reporting benchmark results and a reported 18.9% reduction in total adaptation time for its calibrated version.
Scientific publication4 min read
A computer-vision system for video re-shooting reported stronger camera control than four comparison methods, with visual quality broadly matching Vista4D in the supplied tests.
Scientific publication4 min read
An arXiv preprint in computer vision describes a task-specific pruning pipeline that builds subnetworks and reports lower pruning costs alongside benchmark gains. The study covers six tasks and nine datasets, but its comparisons come without confidence intervals or run-to-run variability estimates.
Scientific publication3 min read
A preprint reports a two-stage way to target identity-related answers in image-capable AI models using forget data at deletion time, while benchmark tests found visual-perception responses remained largely coherent.
Scientific publication5 min read
A preprint reports mixed results for a lightweight residual adapter tested on five decoder-only language models, with higher average accuracy in many pruned-model comparisons but a lower score in one random-layer condition.
Scientific publication4 min read
A preprint tests a whole-heart segmentation system across acquisition sites and finds its combined appearance augmentation and cleanup recipe has the strongest reported Dice scores, especially on CT.
Scientific publication4 min read
The work reviews spatially varying regularisation and illustrates a learned approach on denoising and accelerated MRI, with stronger evidence of noise-specific adaptation in denoising than in MRI.
Scientific publication5 min read
A preprint reports a 75-fold throughput difference between its RefLAM workflow and manual annotation on seven page-validated books. The released material covers main-text and margin lines, while a later phase retained only confidence-100 main-text lines and a downstream test compared two handwriting-recognition models.
Scientific publication4 min read
A preprint finds frozen image models that scored 0.98 to 0.997 on one public white blood cell dataset lost 34% to 72% of macro-F1 in cross-dataset tests.
Scientific publication5 min read
A methods audit found that four automatic audio-visual synchronization metrics often ranked clips differently, depending on the kind of disruption being tested.
Scientific publication4 min read
A computer-vision preprint reports that GLAD ranked first on nine of 10 measures in Real-IAD and all 10 in MANTA-Tiny, but the study reports no uncertainty estimates or production quality-control results.
Scientific publication5 min read
A preprint reports that YOLOEZ, an open-source no-code interface for YOLO-based defect detection, achieved higher recall, F1 score and IoU than a tuned morphological image-processing baseline on tungsten SEM images. The baseline had higher precision and specificity, showing that the tool's advantage did not extend to every measure.
Scientific publication5 min read
A preprint describes a nonlinear 2D wave model that exchanges state information with localized 3D fluid simulation. Numerical tests reported lower wave-height errors, close continuity across the interface and faster runtime, while exposing limits near breaking conditions.
Scientific publication4 min read
A computational chest X-ray study reported higher NIH-to-CheXpert classification scores with BYOL than ImageNet, but low exact-pair retrieval on OpenI and recoverable source proxies.
Scientific publication4 min read
A new open benchmark pairs ground-level and aerial images across 41 European cities, while tests suggest that adding Mapillary data was associated with better in-the-wild localization.
Scientific publication4 min read
A preprint reports higher PointArena scores for PointRL-trained Qwen3.5 models, plus observed gains on RoboSpatial, BLINK and Ref-Adv.
Scientific publication3 min read
A frozen retrieval system led reported metrics on CIRCO, CIRR and FashionIQ, while matched controls favored role-aware dispatch over a shared read.
Scientific publication4 min read