The paper presents the first systematic evaluation of uncertainty quantification (UQ) methods applied to a foundation model for semantic segmentation. By fine‑tuning a lightweight DPT decoder on the pretrained SAM2 encoder, the authors benchmark four UQ approaches—Monte Carlo Dropout, Deep Sub‑Ensemble, Test‑Time Augmentation, and Evidential Deep Learning—across Cityscapes, NYUv2, and two out‑of‑domain settings, comparing segmentation accuracy, calibration, uncertainty quality, and inference time. The results reveal clear trade‑offs between predictive performance, reliability, and computational cost, underscoring both the promise and current limitations of uncertainty‑aware foundation models for real‑world deployment.
By Steven Landgraf, Joceline Hinz, Markus Ulrich
Institutional Books – Enriched Text is a 2025 release that transforms Harvard Library’s 983,004-volume collection (IB‑HL) into a multilingual, annotated dataset. The pipeline normalizes OCR text while preserving metadata, separating endmatter, detecting paragraph language, clustering duplicates, and scoring bits‑per‑byte, all wrapped in HTML‑like annotations. The resulting IB‑HL‑ET contains 217 B tokens across 983,003 volumes and 1.39 B annotated subtopic paragraphs, enabling users to customize output rather than accept a single editorial decision.
Falcon Perception-HD applies reinforcement learning (GRPO) to autoregressive perception models, aligning them directly with precision and recall metrics rather than relying on maximum‑likelihood fine‑tuning. The RL framework introduces reward design for set‑structured outputs and multi‑head sampling control, enabling state‑of‑the‑art performance in very dense scenes (up to 500 objects) and eliminating common issues such as mask repetitions, NMS, and coordinate deduplication. Hybrid self‑annotation pipelines tailored for difficult referring expressions and dense scenes further boost RL training, with improvements observed across all difficulty levels on PBench and SACO‑Gold, and the model preserves object existence knowledge without negative samples.
The article titled "Jigsaw Jeeves: Building a Puzzle Assistant using Computer Vision" provides a conceptual overview and a step‑by‑step walkthrough of a solution approach implemented in Python. It explains how computer vision techniques can be applied to assist users in solving jigsaw puzzles, detailing the design and implementation choices made in the project. The post serves as a practical guide for developers interested in combining vision algorithms with puzzle‑solving applications.
By Chinmay Kakatkar
CamWorldQA introduces the first benchmark for assessing the perceptual quality of camera‑controlled world video generation, featuring 720 videos generated by six methods from 20 source videos across six camera trajectories, each scored by human raters. The paper also presents CWQA, a no‑reference quality assessment network that combines spatial, temporal motion, and optical flow features to predict quality scores. Experiments show CWQA outperforms existing VQA methods on the CamWorldQA dataset.
The paper presents the first systematic evaluation of uncertainty quantification (UQ) methods applied to a foundation model for semantic segmentation. By fine‑tuning a lightweight DPT decoder on the pretrained SAM2 encoder, the authors benchmark four UQ approaches—Monte Carlo Dropout, Deep Sub‑Ensemble, Test‑Time Augmentation, and Evidential Deep Learning—across Cityscapes, NYUv2, and two out‑of‑domain settings. The study compares segmentation accuracy, calibration, uncertainty quality, and inference time, revealing trade‑offs between predictive performance, reliability, and computational cost.
The paper presents a local traditional OCR pipeline that can be iteratively fine‑tuned on both layout and appearance levels of complex historical Sanskrit manuscripts. By adapting to the specific manuscript distribution, the pipeline improves transcription accuracy across subsequent pages, reducing the need for costly human annotation. The authors apply the method to three manuscripts, release a dataset with detailed layout and Unicode annotations in PAGE‑XML format, and benchmark the results against leading multimodal large language models.
The paper introduces Contrastive Dual Gaussian Processes (CDGP), a weakly supervised approach for anomaly segmentation in industrial visual inspection. CDGP models normal and anomaly predictive distributions over dense tokens, using a posterior-dominance statistic that normalizes predictive-mean differences by joint uncertainty to provide spatial evidence and image-level confidence. The method achieves top performance on MVTec AD~2, KSDD2, and VisA datasets without requiring pixel-level annotations or test-time fitting.
The paper introduces Molmo2Fish, an interactive tool that uses a multimodal large language model to correct imperfect fish tracking predictions in sonar datasets. It demonstrates that the system can effectively improve multi‑object tracking performance through natural language guidance, though further enhancements are needed. The authors provide open‑source code and data for replication.
The paper evaluates out‑of‑the‑box object detection models for automatic target detection and recognition (ATD/R) in military settings. Six YOLO variants and two DETR variants were benchmarked on a new military dataset featuring vehicles, occlusions, and small targets, with performance measured in mAP@0.5 and mAP@0.5:0.95 across air‑to‑ground and ground‑to‑ground perspectives. Findings show larger models and DETR-based approaches perform best, fine‑tuning on the VisDrone dataset improves air‑to‑ground and small‑object performance, yet all models still struggle with small targets in air‑to‑ground scenarios.
By Alma M. Liezenga, Lotte Nijskens, Henrik R. Baumann, Stefan Becker, Simon Bensberg, Niccol\`o Camarlinghi, H{\aa}vard R. Eiring, Alexander W. Johnsgaard, Tanel Liiv, Giuseppe Martino, Matteo Marturini, Matthias Rapp, Jan Erik van Woerden, Alexander Wolpert, Hugo J. Kuijf
The paper introduces COMIC, a reference‑aware safety gate designed for multimodal large language models (MLLMs). COMIC detects the operation requested by a user, identifies visual targets through OCR and open‑vocabulary proposals, and evaluates safety on explicit operation‑target pairs, using max‑risk aggregation and quality‑aware routing to decide whether to allow or block a request. Experiments on several open‑source MLLMs and jailbreak benchmarks show that COMIC improves robustness while maintaining benign utility and efficiency.
By Md Abdullahil Oaphy, Anhao Xiang, Zongxing Xie, Huayue Gu, Chenyu Wang, Honghui Xu
The paper introduces OVRSISBench, a unified benchmark for open‑vocabulary remote sensing image segmentation, and evaluates existing OVS/OVRSIS models, uncovering their shortcomings in remote sensing contexts. Leveraging insights from this evaluation, the authors propose RSKT‑Seg, a new framework featuring a Multi‑Directional Cost Map Aggregation module, an Efficient Cost Map Fusion transformer, and a Remote Sensing Knowledge Transfer module. Experiments on the benchmark demonstrate that RSKT‑Seg outperforms strong baselines by +3.8 mIoU and +5.9 mACC while achieving twice the inference speed.
By Bingyu Li, Haocheng Dong, Da Zhang, Zhiyuan Zhao, Junyu Gao, Xuelong Li
The paper investigates a Multi-Scale Spectral Attention Module (MSAM) for hyperspectral image segmentation in autonomous driving. MSAM uses three parallel 1D convolutions with different kernel sizes (1–11) and adaptive feature aggregation, integrated into UNet’s skip connections. Experiments on urban driving datasets show that MSAM improves mIoU by 2.32% and mF1 by 2.88% over baseline UNet-SC while keeping GPU performance competitive, with optimal kernel combinations varying by dataset.
By Imad Ali Shah, Jiarong Li, Tim Brophy, Martin Glavin, Edward Jones, Enda Ward, Brian Deegan
This study presents a clinically relevant framework for evaluating deep neural networks that segment lymphoma lesions in PET/CT images, addressing gaps such as out‑of‑distribution testing and comparison with expert annotators. Using 611 multi‑institutional cases, the authors assess four networks (ResUNet, SegResNet, DynUNet, SwinUNETR) with lesion‑specific metrics, detection criteria, and metabolic‑characteristic‑based thresholds, finding that models perform best on large, intense lesions. The work also demonstrates that network errors mirror those of physicians, highlighting shared challenges with small, faint lesions.
By Shadab Ahamed, Yixi Xu, Sara Kurkowska, Claire Gowdy, Joo H. O, Ingrid Bloise, Don Wilson, Patrick Martineau, Fran\c{c}ois B\'enard, Fereshteh Yousefirizi, Rahul Dodhia, Juan M. Lavista, William B. Weeks, Carlos F. Uribe, Arman Rahmim
Intent-Driven Dynamic Chunking (IDC) segments documents by predicting user queries with a Large Language Model and then applying dynamic programming to find optimal chunk boundaries. This method outperforms traditional fixed-length or coherence-based segmentation on five out of six question-answering datasets, improving top-1 retrieval accuracy by 5% to 67% and reducing the number of chunks by 40–60% while maintaining 93–100% answer coverage. IDC demonstrates that aligning document structure with anticipated information needs can significantly boost retrieval performance for long and heterogeneous documents.
By Christos Koutsiaris
The 10th AI City Challenge, held alongside ECCV 2026, celebrates a decade of benchmarking for intelligent transportation, smart cities, and physical AI. Since its 2017 inception focused on vehicle detection, classification, and tracking, the challenge has expanded into a comprehensive benchmark suite covering multi‑camera perception, multimodal reasoning, synthetic‑to‑real learning, generative forecasting, and privacy‑preserving evaluation. The 2026 edition saw 325 registered teams from 26 countries, with six main tracks—spanning multi‑camera 3D perception, transportation safety captioning and VQA, traffic anomaly reasoning, text‑based person anomaly search, generative traffic video forecasting, and cross‑city object detection—plus two out‑of‑domain leaderboards for fisheye traffic‑violation understanding and pedestrian situated‑intent VQA.
By Zheng Tang, Shuo Wang, David C. Anastasiu, Ming-Ching Chang, Anuj Sharma, Quan Kong, Munkhjargal Gochoo, Jun-Wei Hsieh, Tomasz Kornuta, Zhedong Zheng, Renran Tian, Judah Goldfeder, Fulgencio Navarro, Yuxing Wang, Yizhou Wang, Sameer Satish Pusegaonkar, Anqi Li, Nalin Dadhich, Ridham Kachhadiya, Dhanishtha Patil, Haoquan Liang, Jiajun Li, Han Zhang, Yilin Zhao, Zaid Pervaiz Bhat, Shuyu Yang, Ashutosh Kumar, Rong Wang, Rafael Martin Nieto, Peter Christiansen, Ahmed Abduljawad, Mohanrasu Shanmugam, Nadeem Shaik, Sujit Biswas, Xunlei Wu, Vidya Murali, Rama Chellappa
The paper investigates how confidence‑based model cascades, which route easy inputs to lightweight models and defer difficult cases to larger ones, behave under data perturbations. It shows that static corruptions can either break the routing signal or degrade both models, while sequential perturbations cause predictions to stabilize but deferral to fail, revealing three distinct failure modes. The study highlights that energy‑efficient cascades must be evaluated not only on clean accuracy but also on routing reliability when the input distribution shifts.
By Pallavi Mitra, Jai Kushwaha, Felix Biessmann
The paper introduces an explainable AI framework for automated skill assessment in cataract surgery, leveraging the world’s largest dataset of 2,000 surgical videos. Using advanced computer vision and signal‑processing techniques, the system extracts ten objective motion‑based metrics that correlate strongly with expert subjective ratings, achieving up to 87% accuracy. The framework’s explainability distinguishes it from prior opaque classification tools, offering transparent, quantitative performance indicators that could complement or replace traditional scoring methods.
By Mohammad Javad Ahmadi, Hamid D. Taghirad
The paper introduces a kernel-based target-normalized integration (KTI) method for data collaboration analysis, addressing the limitations of linear obfuscation and integration that can increase reconstruction risk. KTI extends linear target-normalized integration (LTI) by applying kernel ridge regression and an eigenvalue problem, and incorporates graph regularization and centering constraints to embed geometric and target-variable information. Experiments on image classification tasks show that KTI outperforms existing linear methods, especially when combined with target-variable-aware regularization, and highlight how dimensionality reduction choices impact accuracy and reconstruction risk.
By Yamato Suetake, Yuta Kawakami, Shunnosuke Ikeda, Yuichi Takano
The paper presents a method that leverages sparse expert point annotations from historical benthic surveys to improve dense segmentation of marine imagery. By using these points as visual prompts for the SAM2 foundation model and introducing a mechanism to filter out unreliable points, the authors generate high‑quality pseudo‑ground‑truth masks that train more accurate fine‑grained semantic segmentation models. The approach is validated on public benthic datasets and a new benchmark featuring real‑world sparse annotations, aiming to enable scalable ecological analysis.
By Cesar Borja, Breck A. McCollum, Jarret E. Byrnes, Kenneth Sebens, Ana C. Murillo