# object-detection Published articles for object-detection. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Align Your Query: Representation Alignment for Multimodality Medical Object Detection DevFeed: [Align Your Query: Representation Alignment for Multimodality Medical Object Detection]() Original publisher: [Read original article]() Author: Ara Seo, Bryan Sangwoo Kim, Hyungjin Chung, Jong Chul Ye Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Medical imaging](), [Representation learning](), [Embeddings]() Tags: [abstract](), [align](), [alignment](), [annotations](), [approach](), [architectures](), [arxiv](), [attention](), [challenge](), [class](), [context](), [cs-ai](), [cs-cv](), [cs-lg](), [define](), [detection](), [effective](), [effectively](), [medical](), [multimodality](), [object](), [object-detection](), [query]() ### AI overview For medical object detection trained across modalities such as CXR, CT, and MRI, the paper proposes aligning DETR-style object queries with modality context to address heterogeneous data statistics and disjoint representations. MoCA uses self-attention to propagate compact, text-derived modality tokens through the query set; QueryREPA adds a short pretraining stage with a modality-balanced contrastive objective to align queries with those tokens. The abstract reports consistent AP improvements across jointly trained modalities, with minimal latency and no architectural modifications. It does not provide numerical results or further evaluation details. ### Source excerpt arXiv:2510.02789v3 Announce Type: replace-cross Abstract: Medical object detection suffers when a single detector is trained on mixed medical modalities (e.g., CXR, CT, MRI) due to heterogeneous statistics and disjoint representation spaces. To address this challenge, we turn to representation alignment, an approach that has proven effective for bringing features from different sources into a shared space. Specifically, we target the representations of DETR-style object queries and propose a simple, detector-agnostic framework to align them with modality context. First, we define modality tokens: compact, text-derived embeddings encoding imaging modality that are lightweight and require no extra annotations. We integrate the modality tokens into the detection process via Multimodality Context Attention (MoCA), mixing object-query representations via self-attention to propagate modality context within the query set. This preserves DETR-style architectures and adds negligible latency while injecting modality cues into object queries. We further introduce QueryREPA, a short pretraining stage that aligns query representations to their modality tokens using a task-specific contrastive objective with modality-balanced batches. Together, MoCA and QueryREPA produce modality-aware, class-faithful queries that transfer effectively to downstream training. Across diverse modalities trained altogether, the proposed approach consistently improves AP with minimal overhead and no architectural modifications, offering a practical path toward robust multimodality medical object detection. ## Energy-Aware Open-Vocabulary Detection Scheduling for Low-Power Robotic Autonomy DevFeed: [Energy-Aware Open-Vocabulary Detection Scheduling for Low-Power Robotic Autonomy]() Original publisher: [Read original article]() Author: Abdul Basit, Muhammad Shafique Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [scheduling](), [Jetson Orin](), [Sustainable Computing]() Tags: [agilex](), [autonomy](), [cs-ai](), [cs-ro](), [detection](), [embedded](), [energy](), [inference](), [jetson](), [low-power](), [mission](), [mobile](), [navigation](), [object-detection](), [policies](), [power](), [power-management](), [reliability](), [robot](), [robotic](), [runtime](), [scheduling]() ### AI overview A robot-state-aware scheduler reduces open-vocabulary detection workload by switching among low-, mid-, and high-power YOLO-World-S modes using odometry, map changes, and detector confidence. On an AgileX LIMO across 40 navigation trials, it achieved 8/10 mission successes versus 7/10 for each fixed-rate baseline, with 5.2% lower mean Jetson input power and a 34% lower detection call rate than the 8 Hz baseline. It did not minimize Jetson input energy over full missions, since longer runtimes can offset power savings. In a separate 60-second stationary test, it reduced mean power from 8.045 W to 7.301 W and energy from 484.7 J to 438.1 J. The results suggest competitive task reliability alongside workload power savings, while mission energy should be considered with task progress and runtime. ### Source excerpt arXiv:2611.07663v1 Announce Type: cross Abstract: Open-vocabulary object detection (OVD) lets robots search for run-time object queries, but fixed-rate inference spends embedded compute on low-value frames during idle motion, map revisits, or stable viewpoints. This paper presents an energy-aware OVD scheduler for low-power mobile robots. The scheduler uses odometry, map change, and detector confidence to switch among low-, mid-, and high-power YOLO-World-S modes on a Jetson Orin Nano. We evaluate the energy-quality trade-off on an AgileX LIMO using 40 real navigation trials: two target objects, five placements per target-policy pair, and four perception policies. Across bottle and cup trials, the scheduler records 8/10 mission success compared with 7/10 for each fixed-rate baseline, while reducing mean Jetson input power (a Jetson-side power-draw metric) by 5.2% relative to an 8 Hz high-rate baseline and lowering achieved OVD call rate by 34.0%. It does not minimize mission-window Jetson input energy because longer missions can erase instantaneous power savings. A separate 60 s stationary microbenchmark isolates the perception workload: the scheduler lowers mean power from 8.045 W to 7.301 W and energy from 484.7 J to 438.1 J relative to the 8 Hz baseline. These results support robot-state-aware OVD scheduling as a promising power-management mechanism with competitive task reliability, while showing that mission-window Jetson input energy must be read together with task progress and runtime. ## MonoLaDN: Language-Guided DeNoising for Monocular 3D Object Detection DevFeed: [MonoLaDN: Language-Guided DeNoising for Monocular 3D Object Detection]() Original publisher: [Read original article]() Author: Kiet Dang Vu, Cao Le Minh Khoa, Nguyen Viet Tri Pham, An Nguyen Gia, Tam Le Thuc Ho, Duc Dung Nguyen Published: 2026-11-05T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [vlm]() Tags: [3d](), [cs-ai](), [cs-cv](), [denoising](), [detection](), [dimension](), [domains](), [experiments](), [features](), [framework](), [image](), [inference](), [knowledge](), [language](), [language-models](), [latency](), [model](), [models](), [noise](), [object](), [object-detection](), [semantic]() ### AI overview MonoLaDN transfers a vision-language model's geometric priors into a monocular 3D detector during training, while keeping inference vision-only. A frozen VLM uses images and ground-truth boxes to produce structured reasoning about object occlusion, depth, dimensions, and angle. The detector is trained to denoise perturbed reasoning queries and reconstruct the clean reasoning, encouraging its depth-guided features to retain those priors. The reasoning path is removed at inference, so the deployed model adds no latency or parameters. The abstract reports consistent gains over state-of-the-art detectors on KITTI 3D and nuScenes, and generalization to new domains; it provides no specific metrics or domain details. ### Source excerpt arXiv:2611.06282v1 Announce Type: cross Abstract: Monocular 3D object detection offers a low-cost route to 3D perception, but it remains fundamentally ill-posed, since depth must be inferred from a single view. Recent transformer-based detectors learn geometry purely from image features and 3D box annotations, lacking the high-level semantic priors that vision-language models (VLMs) acquire from large-scale pretraining. VLMs offer such priors, but deploying one directly as a detector is impractical due to its heavy inference cost. We propose MonoLaDN, a language-guided denoising framework that transfers the geometric reasoning of a VLM into a monocular 3D detector while keeping the deployed model purely vision-based. During training, a frozen VLM produces structured reasoning about each object's occlusion, depth, dimension, and angle from the image and its ground-truth boxes. We encode this reasoning into queries, perturb them with noise, and train the detector to reconstruct the clean reasoning. Reconstructing the corrupted reasoning forces the detector's depth-guided features to internalize the VLM's geometric priors, so this knowledge remains available even after the reasoning path is discarded at inference, adding no latency or parameters to the deployed model. Experiments on KITTI 3D and nuScenes show that MonoLaDN consistently outperforms state-of-the-art detectors and generalizes well to new domains. ## Detection Instead of Inversion: T1-T2 Microstructure Quantification With Detection Transformers DevFeed: [Detection Instead of Inversion: T1-T2 Microstructure Quantification With Detection Transformers]() Original publisher: [Read original article]() Author: Sebastian Endt, Fatih \\"Ozkan, Marion I. Menzel Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Transformers](), [Neural Network]() Tags: [abstract](), [approach](), [arxiv](), [challenge](), [code](), [correlation](), [cs-cv](), [cs-lg](), [data](), [design](), [detection](), [faster](), [full](), [instead](), [magnitude](), [model](), [object](), [object-detection](), [orders](), [physics-med-ph](), [precision](), [transformers](), [with]() ### AI overview The study replaces voxel-wise inverse Laplace reconstruction of T1-T2 spectra with a Detection Transformer that maps 64 TI/TE signals directly to compartments described by T1, T2, and signal fraction. On simulated data, it reports mAP@10 of 76.8%, compared with 53.8% for the authors' previous model and 66.1% for regularized NNLS. DETR reportedly outperforms NNLS in almost every tested setting across compartment count, signal fraction, and SNR, with reconstruction orders of magnitude faster. The supplied abstract does not establish performance on measured data. ### Source excerpt arXiv:2611.01790v1 Announce Type: cross Abstract: T1-T2 correlation MRI has strong potential to quantify microstructural compartments without imposing a biological model. Traditionally, full T1-T2 spectra per voxel are reconstructed with an ill-conditioned and time-consuming Inverse Laplace Transform. We propose to view this challenge as an object detection task: A Detection Transformer (DETR) takes 64 TI/TE signals and directly predicts a set of compartments specified by T1, T2, and signal fraction. We achieve a mean average precision mAP@10 of 76.8% on simulated data, outperforming both our previous model design (53.8%) and regularized NNLS (66.1%). We show that our DETR approach outperforms NNLS in almost every setting across compartment count, compartmental signal fraction, and SNR. Reconstruction is orders of magnitude faster than NNLS. The results are fully reproducible from our public code: github.com/SebastianEndtTHI/t1t2-detr ## Joint Trajectory, Sensing, and Computation Co-Design for UAV-Assisted Federated Edge Vision DevFeed: [Joint Trajectory, Sensing, and Computation Co-Design for UAV-Assisted Federated Edge Vision]() Original publisher: [Read original article]() Author: Kaichen Liu, Yueling Che, Sheng Luo, Jianhua Tang Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Networking]() Topics: [Transformers](), [Image](), [smartdevices]() Tags: [algorithm](), [battery](), [co-design](), [convergence](), [coupling](), [cs-ni](), [data](), [detection](), [edge](), [federated](), [layer](), [local](), [mobility](), [model](), [object-detection](), [optimization](), [probability](), [sensing](), [simulations](), [time](), [vision]() ### AI overview The paper proposes a four-phase co-design of UAV trajectories, visual sensing, and computation to reduce expected training time for object-detection models under battery constraints. It uses a uniform sensing probability to address data heterogeneity and convert federated convergence requirements into mobility constraints. A two-layer alternating optimization algorithm adapts local data workloads, balancing the number of training rounds against in-place computation delays. Simulations report better convergence and accuracy than the baselines tested; the excerpt provides no further details about those baselines or the size of the improvements. ### Source excerpt arXiv:2611.00790v1 Announce Type: new Abstract: This paper investigates UAV-assisted federated edge vision for training computationally heavy object detection models. Existing studies treat training data as static, ignoring the tight coupling between dynamic visual sensing and massive in place computational burdens. To minimize the expected training time under battery constraints, we propose a four-phase joint trajectory, sensing, and computation co-design. To mitigate data heterogeneity, we use a uniform sensing probability to translate federated convergence requirements into UAV mobility constraints. Furthermore, a two-layer alternating optimization (AO) algorithm is developed to adaptively determine the local data workload, optimally balancing the required training rounds against in-place computation delays. Simulations validate the superiority of our approach in model convergence and accuracy over baselines. ## Agentic LinkDreamer: Adaptive Neuro-Fuzzy World-Model Decision Making for Intermittent V2X Cooperative Perception DevFeed: [Agentic LinkDreamer: Adaptive Neuro-Fuzzy World-Model Decision Making for Intermittent V2X Cooperative Perception]() Original publisher: [Read original article]() Author: Zhaoyu Liu, Dongmin Wang, Liu Cao, Jijing Cai, Kai Fang, Syed Ishtiaque Ahmed, Almetwally M. Mostafa Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Networking]() Topics: [World models](), [decision-making](), [Physical AI](), [Artificial Intelligence]() Tags: [adaptive](), [agentic](), [budget](), [communication](), [controller](), [cs-ni](), [decision-making](), [detection](), [framework](), [latency](), [lidar](), [link](), [memory](), [model](), [motion](), [object-detection](), [outage](), [outages](), [overhead](), [world](), [world-model]() ### AI overview Agentic LinkDreamer combines a recurrent world model with an interpretable neuro-fuzzy controller to manage cooperative perception when V2X links fail. It predicts missing bird's-eye-view features and chooses among ego-only perception, motion-compensated history reuse, latent imagination, low-rate summaries, or high-rate feature refresh. Decisions use prediction uncertainty, collaborator complementarity, memory freshness, link quality, and remaining communication budget; after link recovery, received features help recalibrate uncertainty and update memory. The evaluation uses a geometry-derived degradation protocol on V2X-Real trajectories and LiDAR scene geometry. On the V2V split, the system reports 90.95 AP@0.5 on outage frames and recovery of 92.9% of the cooperative-perception performance gap. At a high-performance operating point, it is 0.65 AP below the cooperative upper reference while using about 31% less communication than continuous high-rate cooperation. ### Source excerpt arXiv:2610.39258v1 Announce Type: new Abstract: Intermittent vehicle-to-everything (V2X) communication can remove collaborator information needed for cooperative perception, degrading object detection during link outages. Existing recovery methods reconstruct missing features but rarely determine when to trust predictions, reuse historical observations, or request fresh information. This paper proposes Agentic LinkDreamer, a soft-computing framework integrating a latent world model with an interpretable neuro-fuzzy decision agent. It maintains causal collaborator memory and uses a relative-motion-conditioned recurrent world model to predict unavailable bird's-eye-view (BEV) features. Prediction uncertainty, perceptual complementarity, memory freshness, communication quality, and residual communication budget form a compact decision state. A differentiable Takagi-Sugeno controller selects among five modes: ego-only perception, motion-compensated historical feature reuse, latent imagination, low-rate temporal-summary refresh, and high-rate collaborator-feature refresh. The policy balances detection risk, communication overhead, latency, predictive uncertainty, and action switching. After link recovery, reflection compares predicted and received features to update uncertainty calibration and causal memory. For reproducible evaluation without assuming unavailable physical-layer traces, we construct a geometry-derived V2X degradation protocol using recorded trajectories and LiDAR-derived scene geometry from V2X-Real. On the real-world V2X-Real V2V split, Agentic LinkDreamer achieves 90.95 AP@0.5 on outage frames and recovers 92.9% of the cooperative-perception performance gap. At a high-performance operating point, it remains only 0.65 AP below the cooperative upper reference while reducing communication volume by approximately 31% relative to continuous high-rate cooperation. This improves resilience under V2X outages. ## LiDAR-SHIELD: A Fully Dockerized Adversarial Robustness Assessment Framework for LiDAR-Based 3D Perception Models DevFeed: [LiDAR-SHIELD: A Fully Dockerized Adversarial Robustness Assessment Framework for LiDAR-Based 3D Perception Models]() Original publisher: [Read original article]() Author: Alexandros Gkillas, Ioulia Kapsali, Christos Anagnostopoulos, Aris S. Lalos Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Adversarial attacks]() Tags: [3d](), [adapter](), [adversarial](), [assessment](), [attacks](), [autonomous](), [autonomous-driving](), [cs-ai](), [cs-cv](), [dataset](), [detection](), [framework](), [gradient](), [http](), [info](), [lidar](), [loader](), [model](), [models](), [object-detection](), [robotics](), [robustness]() ### AI overview LiDAR-SHIELD aims to simplify adversarial robustness testing across heterogeneous LiDAR models by keeping target models outside its Dockerized toolkit. A model is exposed through three HTTP endpoints--/info, /forward, and /gradient--and connecting a model or dataset requires a loader function or adapter. The shared engine-service protocol supports 3D semantic segmentation and 3D object detection, with PGD and Square Attack for segmentation, and point perturbation, removal, injection, and LiDAttack for detection. The abstract describes the framework as a way to systematically expose corner cases and vulnerabilities; it provides no comparative evaluation results. ### Source excerpt arXiv:2610.39848v1 Announce Type: cross Abstract: LiDAR-based 3D perception underpins autonomous driving and robotics, yet deploying these models safely requires stress-testing them against adversarial inputs before real-world use. Although state-of-the-art adversarial attacks exist, their implementations are often fragmented across individual repositories, each tied to a specific architecture or task and requiring manual environment reconstruction and model-specific integration code. We present LIDARSHIELD, a unified and fully Dockerized framework that reverses this workflow: rather than installing the target model inside the toolkit, LIDARSHIELD exposes any model through three lightweight HTTP endpoints, namely /info, /forward, and /gradient As a result, connecting a new model or dataset requires only a single loader function or adapter. The framework supports both 3D semantic segmentation and 3D object detection through the same engine-service protocol, integrating a task-aware suite of state-of-the-art attacks, including PGD and Square Attack for segmentation, and point perturbation, point removal, point injection, and LiDAttack for detection. This enables systematic exposure of corner cases and vulnerabilities across heterogeneous LiDAR backbones. The LiDAR-Shield codebase is publicly available at https://github.com/alexandrosgk/LiDAR-shield. ## Smart Surveillance: Deep Learning and Hybrid Approaches for Dynamic Scene Analysis, Object Detection and Motion Tracking DevFeed: [Smart Surveillance: Deep Learning and Hybrid Approaches for Dynamic Scene Analysis, Object Detection and Motion Tracking]() Original publisher: [Read original article]() Author: Shahran Rahman Alve Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Video Analytics](), [recording](), [image recognition](), [quantization]() Tags: [abstract](), [analysis](), [arxiv](), [association](), [background](), [based](), [benchmark](), [but](), [cs-ai](), [cs-cv](), [deep](), [deep-learning](), [detection](), [hybrid](), [learning](), [motion](), [motion-tracking](), [object](), [object-detection](), [smart](), [surveillance](), [tracking]() ### AI overview In the seven-scene benchmark, detector-based recording triggers stored fewer idle frames than the six motion triggers (19-28% versus 36-89%) and remained below 29% under simulated camera jitter, when several motion methods stored nearly all idle frames. Detecting every eighth frame retained 99.7% of event frames at 9.8 ms CPU per frame and was cheaper and more robust than motion-gated detection. On VisDrone, increasing YOLO11n input resolution improved AP from 15.5 to 26.8; INT8 reduced latency 17% for a 0.3 AP cost. Longer tracking memory improved IDF1, but trackers recovered only 39-44% of overlapped people, with detection identified as the limiting failure. ### Source excerpt arXiv:2412.05331v4 Announce Type: replace-cross Abstract: Event-triggered recording reduces the footage a surveillance camera must store, but the trigger decides how much idle video is kept, how many events are missed, and how much computation is spent. We compare six motion triggers (frame differencing, dense optical flow, MOG2, KNN, CNT, GSOC), a CNN detector, strided detection, and three motion-gated hybrids under one recording policy, on a benchmark of seven static-camera scenes (5,800 frames, 27 annotated events) with thresholds transferred to unseen scenes by leave-one-video-out. Detector-based triggers stored 19-28% of idle frames, against 36-89% for motion triggers. Under simulated camera jitter, frame differencing, optical flow, MOG2 and KNN stored 97-100% of idle frames, while detector triggers stayed below 29%. Running the detector on every 8th frame kept 99.7% of event frames at 9.8 ms of CPU per frame, cheaper and more robust than gating it with background subtraction. On VisDrone, raising YOLO11n's input from 640 to 1280 px raised AP from 15.5 to 26.8, more than a larger model, and INT8 quantization cut latency by 17% for 0.3 AP. In tracking, longer track memory raised IDF1 from 70.5 to 76.7, but every tracker recovered only 39-44% of overlapped people, because detection rather than association failed. ## How Interfaze serves 50 million requests a month on Vercel DevFeed: [How Interfaze serves 50 million requests a month on Vercel]() Original publisher: [Read original article]() Author: Kevin Sundstrom Published: 2026-10-23T04:00:00Z Content type: article Language: en Sources: [Vercel News]() Topics: [Vercel](), [Serverless](), [Back end](), [AI Infrastructure](), [CI/CD](), [Large Language Model](), [document ai]() Tags: [ai](), [ai-gateway](), [api](), [backend](), [classification](), [data-preprocessing](), [fallback](), [fluid](), [functions](), [object-detection](), [ocr](), [routing](), [sandbox](), [scalability](), [scraping](), [speech-to-text](), [third-party](), [vercel]() ### AI overview Interfaze's models target tasks such as document processing, OCR, scraping, speech recognition, and classification where results need to be verifiable. For example, invoice extraction returns a value with its page location and a confidence score. The company says its API serves 50 million requests a month, with its frontend and backend running as serverless functions in one codebase; Vercel Sandbox runs generated code in isolation. Fluid compute scales with traffic and supports Puppeteer scraping within Vercel's memory and timeout limits. For data cleaning, AI Gateway automatically routes requests to another provider when one fails or throttles, and lets Interfaze compare providers' performance even when they claim to host the same open source model. Interfaze also uses Workflow SDK for usage-based email sequences, with automatic retries and workflows that can resume months later with state intact. Preview deployments give model versions distinct URLs and logs, and let customers test a version before release; the Vercel portion of a release reportedly builds in about three minutes, while GPU weight hosting takes longer. ### Source excerpt Interfaze on Vercel 50 million requests a month Backend running on Fluid compute Multiple model providers behind one fallback layer with AI Gateway Interfaze builds models for deterministic work that general-purpose LLMs handle poorly: document processing, OCR, object detection, web scraping, speech to text, and classification. As founder Yoeven Khemlani puts it, "Most models aren't deterministic by nature. That's fine when you're drafting an essay. But when a hospital or a legal team needs to process documents, you don't reach for an LLM." Ask an LLM for the total on an invoice and you get a number with no easy way to check where it came from. Interfaze's models return the number along with a bounding box showing where on the page it came from and a confidence score for how sure the model is. Every result can be checked, and the models still take prompts the way an LLM does. Everything around Interfaze's models runs on Vercel, including the backend. That spans the API serving 50 million requests a month, the sandbox that executes code the model writes, and the routing that keeps data preprocessing running when a third-party model provider goes down. The backend required no infrastructure work Interfaze's frontend and backend both run on Vercel. The backend handles every request to Interfaze's API, whether a customer is calling the model or uploading a file. All of it runs as serverless functions in the same codebase as the frontend. When a request needs computation, the model hands the code to Vercel Sandbox, which runs it in isolation. There was no separate infrastructure to set up. Fluid compute handles the heavy jobs Scalability was Yoeven's top requirement. The goal is serving the model to millions of users, and Fluid compute scales with traffic on its own. Heavier work runs there too. Vercel's memory and timeout limits are high enough that Interfaze's web scraping, which runs on Puppeteer, works out of the box. AI Gateway makes model routing reliable at scale ## Contrastive Hypergraph Source-free Domain Adaptive Object Detection in Adverse Weather Conditions--Extended Version DevFeed: [Contrastive Hypergraph Source-free Domain Adaptive Object Detection in Adverse Weather Conditions--Extended Version]() Original publisher: [Read original article]() Author: Jicheng Yuan, Tung Kieu, Manh Nguyen-Duc, Manfred Hauswirth, Danh Le-Phuoc Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Representation learning](), [Embeddings](), [Image processing]() Tags: [abstract](), [adaptive](), [art](), [arxiv](), [based](), [benchmarks](), [capture](), [confidence](), [context](), [cs-ai](), [cs-cv](), [cs-lg](), [data](), [detection](), [domain](), [effectively](), [encoding](), [free](), [object](), [object-detection](), [source](), [version](), [weather]() ### AI overview REUNION adapts object detectors to fog, rain, snow, and low-light target domains when the original training data are unavailable. It builds a hypergraph over target-domain object proposals, connecting them through image context, weather-aware groups, semantic prototypes, and confidence-aware links between reliable and uncertain instances. Its group-wise Hyper-InfoNCE objective trains representations across these groups so information from confident proposals can support corrupted or low-contrast ones. The abstract reports improved domain-shift handling and state-of-the-art results across diverse benchmarks, but provides no benchmark names or quantitative results. ### Source excerpt arXiv:2610.24532v1 Announce Type: cross Abstract: Adverse weather conditions such as fog, rain, snow, and low illumination introduce structured visibility degradations that severely impair object detectors trained on clear-weather data. When source data are inaccessible due to privacy or trans- mission constraints, source-free domain adaptation (SFDA) becomes necessary. However, existing SFDA methods based on pairwise contrastive learning struggle to capture high-order semantics and often suffer from agreement collapse under severe corruption. We propose REUNION, a hypergraph-guided SFDA framework for object detection in adverse weather conditions. REUNION constructs a contrastive hypergraph over target-domain object proposals, encoding high-order relations through intra-image context, weather-aware grouping, prototype-based semantic anchors, and uncertainty-aware connections between reliable and low-confidence instances. To effectively exploit these structured relations, we introduce a group-wise Hyper-InfoNCE objective that optimizes representations at the hyperedge level, enabling semantic information to propagate from confident proposals to corrupted or low-contrast instances. Experiments with diverse benchmarks demonstrate that REUNION effectively mitigates domain shifts and achieves state-of-the-art performance in SFDA settings. ## Google Is Back! Gemini 4 Argon Leads Our Vision Benchmark DevFeed: [Google Is Back! Gemini 4 Argon Leads Our Vision Benchmark]() Original publisher: [Read original article]() Author: Piotr Skalski Published: 2026-10-15T13:59:33Z Content type: comparison Language: en Sources: [Roboflow]() Topics: [smolvlm](), [object-detection](), [Computer vision](), [vlm]() Tags: [benchmark](), [cost](), [detection](), [gemini](), [gemini-4-argon](), [google](), [models](), [multimodal](), [object](), [object-detection](), [ocr](), [speed](), [video](), [vision](), [vision-language-models]() ### AI overview The article evaluates Gemini 4 Argon on object detection, segmentation, OCR, and video tasks, comparing its accuracy, speed, and cost with other vision-language models. It reports that Argon leads the publisher's overall vision benchmark and describes trade-offs, including slower response times alongside lower estimated costs than Astra in the tested workload. ### Source excerpt We evaluate Gemini 4 Argon across object detection, segmentation, OCR, and video tasks, comparing its accuracy, speed, and cost with other leading vision-language models. ## Introducing Roboflow Labs: Advancing the Frontier of Real World Capabilities DevFeed: [Introducing Roboflow Labs: Advancing the Frontier of Real World Capabilities]() Original publisher: [Read original article]() Author: Jake Seaton Published: 2026-09-30T15:23:29Z Content type: release Language: en Sources: [Roboflow]() Topics: [Computer vision](), [Physical AI](), [Frontier AI](), [World models]() Tags: [ai](), [artificial-intelligence](), [benchmarks](), [environments](), [evals](), [labs](), [object-detection](), [ocr]() ### AI overview Roboflow announces Labs, an applied research division focused on advancing real-world AI capabilities. It will publish research and open models for edge deployment, collaborate with frontier labs on model evaluation and training data, and develop enterprise benchmarks for vision tasks. ### Source excerpt Advancing the Frontier of Real World Capabilities ## From Zero to Edge Intelligence: A TinyML Workshop at CET DevFeed: [From Zero to Edge Intelligence: A TinyML Workshop at CET]() Original publisher: [Read original article]() Author: Abhinav Published: 2026-09-28T08:06:53Z Content type: article Language: en Sources: [Seeed Studio]() Topics: [tinyml](), [Machine learning](), [Physical AI]() Tags: [arduino](), [community](), [computer-vision](), [edge](), [edge-ai](), [edge-impulse](), [esp32s3](), [events](), [grove](), [image-classification](), [innovation](), [iot](), [keyword-spotting](), [machine-learning](), [news](), [object-detection](), [seeed-ranger](), [seeed-studio-xiao](), [sensecraft](), [tinyml](), [workshop](), [xiao]() ### AI overview A two-day TinyML workshop at CET introduced 56 participants to edge AI through hands-on model training and deployment. Participants built image classification and object detection models on Seeed hardware, then used Edge Impulse to train audio models for keyword spotting, thunder detection, and animal detection. ### Source excerpt By the second afternoon, students at CET had built AI that could listen for a The post From Zero to Edge Intelligence: A TinyML Workshop at CET appeared first on Latest News from Seeed Studio. ## The ATHER workshop introduced students to TinyML through hands-on model building DevFeed: [The ATHER workshop introduced students to TinyML through hands-on model building]() Original publisher: [Read original article]() Author: Abhinav Published: 2026-09-28T07:19:08Z Content type: article Language: en Sources: [Seeed Studio]() Topics: [tinyml](), [Machine learning](), [AI Development](), [Physical AI]() Tags: [arduino](), [chip](), [community](), [data](), [edge-ai](), [edge-impulse](), [events](), [grove](), [innovation](), [iot](), [mace](), [machine-learning](), [news](), [object-detection](), [seeed-ranger](), [sensecraft](), [tinyml](), [workshop](), [xiao]() ### AI overview The ATHER full-day workshop introduced 72 students to TinyML, which runs machine-learning models directly on low-power devices without constant internet access. Participants used Seeed Studio's SenseCraft AI platform and hardware to build image classification and object detection models, then collected data, trained custom models with Edge Impulse, and deployed them to devices. ### Source excerpt We usually picture AI as something massive that runs in the cloud. On September 12, The post Tiny Models, Big Curiosity: The ATHER TinyML Workshop at MACE appeared first on Latest News from Seeed Studio. ## Compact Long-Wave Infrared camera pairs STM32N6 MCU with Lynred ATI320 thermal sensor DevFeed: [Compact Long-Wave Infrared camera pairs STM32N6 MCU with Lynred ATI320 thermal sensor]() Original publisher: [Read original article]() Author: Debashis Das Published: 2026-09-25T10:00:21Z Content type: news Language: en Sources: [CNX Software - Embedded Systems News]() Topics: [Embedded Systems](), [Microcontroller](), [Hardware](), [Artificial Intelligence](), [object-detection](), [Inference](), [Image processing](), [real-time]() Tags: [8devices](), [ai](), [ai-inference](), [camera](), [computer-vision](), [drone](), [edge-ai](), [embedded-systems](), [hardware](), [image-processing](), [industrial](), [infrared](), [iot](), [machine-vision](), [mcu](), [object-detection](), [real-time](), [sensor](), [stm32](), [stmicro-stm32]() ### AI overview The 8devices 8Sight T100 is a compact thermal camera for UAS, robotics, and industrial applications. It combines a Lynred ATI320 LWIR sensor with an STM32N6 microcontroller, Neural-ART NPU, and hardware ISP for offline edge AI inference and real-time thermal image processing. The camera provides 320x240 video at 60 fps and supports object detection, anomaly detection, and temperature-triggered events. ### Source excerpt 8devices 8Sight T100 is a compact Long-Wave Infrared (thermal imaging) camera for UAS, robotics, and industrial applications. It combines a Lynred ATI320 LWIR sensor with an STM32N6 Arm Cortex-M55 microcontroller featuring a Neural-ART NPU and a hardware ISP for edge AI inference and real-time thermal image processing. The camera delivers 320x240 thermal video at 60 fps and operates completely offline, allowing you to deploy custom AI models for object detection, anomaly detection, or temperature-triggered events without relying on cloud resources. 8devices 8Sight T100 specifications: MCU - STMicro STM32N657X0H3Q with an Arm Cortex-M55 core up to 800MHz with a 600 GOPS Neural-ART Accelerator (NPU) and hardware ISP System Memory - 256 Mbit (32 MB) RAM @ 200 MHz (16-bit) Storage - 512 Mbit (64 MB) NOR flash Thermal Camera - Lynred ATI320 LWIR (Long-Wave Infrared) sensor Spectral range - 8 to 14 µm Resolution - 320 x 240 pixels Frame rate - [...] The post Compact Long-Wave Infrared camera pairs STM32N6 MCU with Lynred ATI320 thermal sensor appeared first on CNX Software - Embedded Systems News. ## Balancing Bandwidth, Range, and Power in Intelligent Buildings DevFeed: [Balancing Bandwidth, Range, and Power in Intelligent Buildings]() Original publisher: [Read original article]() Author: Andy McFarlane Published: 2026-09-25T08:00:34Z Content type: article Language: en Sources: [EE Times]() Topics: [Internet of things](), [Artificial Intelligence](), [Network](), [Wi-Fi](), [object-detection](), [real-time]() Tags: [ai](), [analog](), [analysis](), [blog](), [business](), [cloud](), [community](), [designline](), [digital](), [edge-ai](), [education](), [ee-times](), [ee-times-university](), [eelife](), [eelive-electronic-design](), [eet](), [electrical-engineering](), [electronic-components](), [electronic-engineering-times](), [electronics](), [embedded-systems](), [firmware](), [fundamentals-courses](), [iot](), [learning](), [mixed-signal](), [network](), [news](), [object-detection](), [opinion](), [part-search](), [product-search](), [semiconductors](), [smart-buildings](), [software](), [tech-papers](), [technology](), [uncategorized](), [video](), [wi-fi](), [wi-fi-halow](), [wireless-and-networking]() ### AI overview This article explains how edge AI in intelligent buildings changes network requirements. It discusses balancing bandwidth, range, and power for devices that process data locally, including cameras, sensors, controllers, and building safety systems. ### Source excerpt As edge AI transforms smart buildings, Wi-Fi HaLow bridges the gap between bandwidth, long range, and low power. The post Balancing Bandwidth, Range, and Power in Intelligent Buildings appeared first on EE Times. ## Introduction to Triton Java API DevFeed: [Introduction to Triton Java API]() Original publisher: [Read original article]() Author: Nikhil Bhargav Published: 2026-09-23T06:48:14Z Content type: tutorial Language: en Sources: [Baeldung]() Topics: [Java](), [model-serving](), [object-detection](), [Inference](), [Deployment](), [Docker](), [gRPC](), [HTTP](), [onnx](), [TensorRT](), [PyTorch](), [Tensorflow]() Tags: [api](), [artificial-intelligence](), [artificial-intelligence-spring-ai-chatclient](), [deployment](), [docker](), [grpc](), [http](), [inference](), [java](), [nvidia](), [object-detection](), [popular](), [server](), [spring-ai-chatclient](), [triton-inference-server]() ### AI overview A tutorial on using NVIDIA Triton Inference Server from Java for machine learning model serving and object detection. It covers the in-process Java API, the gRPC/HTTP client API, Docker setup, and executing a pre-trained YOLO model. ### Source excerpt Learn to use NVIDIA's Triton Inference Server in Java to do object detection in images. The post Introduction to Triton Java API first appeared on Baeldung. ## Roboflow Announces Lower Serverless API Pricing and a Free Core Plan DevFeed: [Roboflow Announces Lower Serverless API Pricing and a Free Core Plan]() Original publisher: [Read original article]() Author: Leo Ueno Published: 2026-09-18T20:17:00Z Content type: release Language: en Sources: [Roboflow]() Topics: [Artificial Intelligence](), [Inference](), [model-deployment](), [object-detection](), [Cloud](), [Serverless]() Tags: [ai](), [cloud](), [deploy](), [developers](), [evaluation](), [inference](), [object-detection](), [pricing](), [product-updates]() ### AI overview Roboflow announces changes to its vision AI platform, including per-image Serverless API pricing, a free Core Plan tier, self-hosted inference, and downloadable model weights. The company says the new pricing is more predictable and cheaper for many workloads. ### Source excerpt Our mission is to make the world programmable, and accelerate vision AI into the world. That means building an ecosystem where developers can experiment, founders can turn prototypes into businesses, and people who know a problem firsthand can use vision AI to solve it. ## GPT-6 Astra Is the Best Vision Model We Have Tested DevFeed: [GPT-6 Astra Is the Best Vision Model We Have Tested]() Original publisher: [Read original article]() Author: Piotr Skalski Published: 2026-09-18T12:50:17Z Content type: comparison Language: en Sources: [Roboflow]() Topics: [gpt-6-astra](), [Computer vision](), [object-detection](), [Benchmark]() Tags: [2026](), [benchmark](), [comparisons](), [computer-vision](), [detection](), [gpt-6](), [object-detection](), [results](), [vision]() ### AI overview Roboflow evaluates GPT-6 Astra on object detection, segmentation, counting, visual reasoning, video re-identification, and robot control. The article reports strong benchmark results and flexible visual prompting, while noting trade-offs in price, latency, and segmentation boundary accuracy. It recommends testing cheaper models first and using Astra where its results reduce enough manual correction to justify the cost. ### Source excerpt We test GPT-6 Astra on object detection, segmentation, counting, visual reasoning, and video, with examples, benchmark results, and cost comparisons. ## GPT-6 Astra Is the Best Vision Model We Have Tested DevFeed: [GPT-6 Astra Is the Best Vision Model We Have Tested]() Original publisher: [Read original article]() Author: Piotr Skalski Published: 2026-09-18T12:50:17Z Content type: comparison Language: en Sources: [Roboflow]() Topics: [Computer vision](), [object-detection](), [vlm](), [Benchmark](), [Prompt Engineering](), [JSON]() Tags: [benchmark](), [comparisons](), [computer-vision](), [detection](), [gpt-6-astra](), [model-evaluation](), [multimodal](), [object-detection](), [openai](), [vision]() ### AI overview Roboflow evaluates GPT-6 Astra on computer vision tasks including object detection, segmentation, counting, visual reasoning, and video. The article reports benchmark results, comparisons with other models, examples of prompting, and Astra's use in Roboflow Auto Annotate. ### Source excerpt We test GPT-6 Astra on object detection, segmentation, counting, visual reasoning, and video, with examples, benchmark results, and cost comparisons. ## GPT-6 Astra for Segmentation DevFeed: [GPT-6 Astra for Segmentation]() Original publisher: [Read original article]() Author: Erik Kokalj Published: 2026-09-17T18:45:11Z Content type: article Language: en Sources: [Roboflow]() Topics: [gpt-6-astra](), [object-detection](), [OpenAI]() Tags: [computer-vision](), [gpt-6-astra](), [json](), [json-schema](), [model-evaluation](), [object-detection](), [openai](), [schema]() ### AI overview The post evaluates GPT-6 Astra for object segmentation. It explains how to prompt the model to return polygon outlines, describes their limitations compared with dense masks, and presents a workflow that combines Astra's object detection with SAM 3 for more accurate segmentation. ### Source excerpt GPT-6 Astra can return polygon outlines for any object you describe. They are not pixel masks, so we benchmarked them against SAM 3 and show how to combine the two. ## Enhancing industrial safety AI with synthetic data on Amazon SageMaker AI DevFeed: [Enhancing industrial safety AI with synthetic data on Amazon SageMaker AI]() Original publisher: [Read original article]() Author: Dimitri Voytan Published: 2026-09-17T15:28:08Z Content type: tutorial Language: en Sources: [Artificial Intelligence]() Topics: [Artificial Intelligence](), [synthetic-data](), [Amazon SageMaker](), [data augmentation](), [object-detection](), [Person Detection](), [Computer vision](), [Training AI Models](), [data-processing](), [Deployment]() Tags: [advanced-300](), [ai](), [amazon-sagemaker-ai](), [computer-vision](), [data-augmentation](), [industrial](), [object-detection](), [person-detection](), [pipeline](), [safety](), [synthetic-data](), [technical-how-to](), [training](), [up]() ### AI overview This tutorial shows how to build a synthetic data augmentation pipeline with Amazon SageMaker AI and Amazon Rekognition for industrial safety computer vision. The pipeline generates photo-realistic, automatically labeled training images for rare and hazardous scenarios near heavy machinery, with experiments reporting up to a 160% improvement in person-detection mAP50 without manual annotation or hazardous photography sessions. ### Source excerpt Learn how to build a synthetic data augmentation pipeline on Amazon SageMaker AI and Amazon Rekognition that generates photo-realistic, auto-labeled training images for industrial safety AI. This approach improved person detection by up to 160% without manual annotation or hazardous data collection near heavy machinery. ## Quiz: Traditional Face Detection With Python DevFeed: [Quiz: Traditional Face Detection With Python]() Original publisher: [Read original article]() Author: Real Python Published: 2026-09-13T12:00:00Z Content type: article Language: en Sources: [Real Python]() Topics: [Python](), [object-detection](), [Algorithms](), [real-time]() Tags: [article](), [object-detection](), [python](), [real-time](), [tutorial]() ### AI overview An interactive 10-question quiz that tests understanding of traditional face detection with Python, including image representation, Haar-like features, integral images, AdaBoost, and cascading classifiers. ### Source excerpt Test your understanding of face detection with Python. Review Haar-like features, integral images, AdaBoost, and cascading classifiers. ## NeoEyes NE302 - A tiny USB-C-powered WiFi 6 Edge AI Vision camera based on STM32N6 MCU DevFeed: [NeoEyes NE302 - A tiny USB-C-powered WiFi 6 Edge AI Vision camera based on STM32N6 MCU]() Original publisher: [Read original article]() Author: Jean-Luc Aufranc (CNXSoft) Published: 2026-09-12T02:19:36Z Content type: news Language: en Sources: [CNX Software - Embedded Systems News]() Topics: [Embedded Systems](), [Microcontroller](), [Wi-Fi](), [Hardware](), [object-detection]() Tags: [c-c-plus-plus](), [camera](), [computer-vision](), [edge-ai](), [embedded-systems](), [hardware](), [mcu](), [object-detection](), [open-source](), [stm32](), [stmicro-stm32](), [vision](), [wifi-6]() ### AI overview The article reports on CamThink's NeoEyes NE302, a compact USB-C-powered Wi-Fi 6 edge AI vision camera built around the STM32N6 MCU. It describes the device's reduced memory and feature set compared with the NE301, along with its camera, wireless, debugging, video-encoding, object-detection, networking, OTA, and security capabilities. ### Source excerpt CamThink NeoEyes NE302 is a tiny WiFi 6 Edge AI camera based on an STM32N6 Arm Cortex-M55 MCU with Neural-ART NPU which the company says is "especially suitable for developers and makers". I initially thought about it as a smaller version of the CamThink NeoEyes NE301 based on the same STM32N6 MCU and 4MP OS04C10 camera sensor. But it's quite a different device. First, it doesn't have a battery, and the USB-C is only used for power. Memory and storage capacities have been reduced to 32 MB PSRAM and 64 MP SPI flash, and a range of built-in and optional features have been removed, including 4G LTE module (global or US), audio wafers, PIR motion connector, and a 16-pin GPIO header, as well as support for PoE power. CamThink NeoEyes NE302 specifications: Vision MCU STMicro STM32N6 MCU Core - Arm 32-bit Cortex-M55 CPU @ up to 800MHz with Arm Helium [...] The post NeoEyes NE302 - A tiny USB-C-powered WiFi 6 Edge AI Vision camera based on STM32N6 MCU appeared first on CNX Software - Embedded Systems News. [Next page]()