New papers from Lorenzo Vaquero on computer vision, multi-object tracking, human–object interaction, vision–language models, and AI explainability.https://vaquero.me/en-ushttps://vaquero.me/publications/2026-cvpr-from-weights-to-concepts/https://vaquero.me/publications/2026-cvpr-from-weights-to-concepts/A data-free, training-free framework that directly analyses CLIP's vision transformer in weight space, decomposing each attention head into singular vectors linked to textual concepts. (IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026)Thu, 01 Jan 2026 00:00:00 GMTVision-Language ModelsAI ExplainabilityFrancesco Gentile, Nicola Dall'Asen, Francesco Tonini, Massimiliano Mancini, Lorenzo Vaquero, Elisa Riccihttps://vaquero.me/publications/2026-perconai-efficient-portrait-segmentation-embedded-devices/https://vaquero.me/publications/2026-perconai-efficient-portrait-segmentation-embedded-devices/SINergy is a scalable portrait segmentation network designed around embedded hardware constraints, reaching real-time segmentation on microcontroller-scale devices. (IEEE International Conference on Pervasive Computing and Communications Workshops, 2026)Thu, 01 Jan 2026 00:00:00 GMTEmbedded VisionEfficient SegmentationRiccardo Benevelli, Alberto Ancilotto, Lorenzo Vaquero, Elisa Ricci, Elisabetta Farellahttps://vaquero.me/publications/2026-acmmm-finehoi/https://vaquero.me/publications/2026-acmmm-finehoi/FineHOI detects unseen human-object interactions by reasoning over dense, part-level visual cues, improving fine-grained recognition without LLM or detector features. (ACM International Conference on Multimedia, 2026)Thu, 01 Jan 2026 00:00:00 GMTHuman-Object InteractionVision-Language ModelsFrancesco Tonini, Lorenzo Vaquero, Mohammad Mahdi Derakhshani, Cees G. M. Snoek, Elisa Ricci, Cigdem Beyanhttps://vaquero.me/publications/2026-nature-humanitys-last-exam/https://vaquero.me/publications/2026-nature-humanitys-last-exam/Humanity's Last Exam is a multimodal benchmark of 2,500 expert-level closed-ended academic questions designed to measure frontier AI capabilities beyond saturated benchmarks. (Nature, 2026)Thu, 01 Jan 2026 00:00:00 GMTBenchmarksAI EvaluationCenter for AI Safety, Scale AI, HLE Contributors Consortiumhttps://vaquero.me/publications/2026-caepia-keynote-lost-and-found/https://vaquero.me/publications/2026-caepia-keynote-lost-and-found/This keynote paper summarizes BUSCA, a work published in ECCV'24. BUSCA plugs into online tracking-by-detection systems to recover objects missed by detectors through proposal generation and decision-transformer association. (Conference of the Spanish Association for Artificial Intelligence, 2026)Thu, 01 Jan 2026 00:00:00 GMTMulti-Object TrackingOnline TrackingLorenzo Vaquero, Yihong Xu, Xavier Alameda-Pineda, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2026-caepia-keynote-superpowering-open-vocabulary-object-detectors-x-ray-vision/https://vaquero.me/publications/2026-caepia-keynote-superpowering-open-vocabulary-object-detectors-x-ray-vision/This keynote paper summarizes RAXO, a work published in ICCV'25. RAXO adapts RGB open-vocabulary detectors to X-ray imagery with training-free visual descriptors and introduces DET-COMPASS for large-scale X-ray OvOD evaluation. (Conference of the Spanish Association for Artificial Intelligence, 2026)Thu, 01 Jan 2026 00:00:00 GMTOpen-Vocabulary DetectionX-ray VisionPablo Garcia-Fernandez, Lorenzo Vaquero, Mingxuan Liu, Feng Xue, Daniel Cores, Nicu Sebe, Manuel Mucientes, Elisa Riccihttps://vaquero.me/publications/2026-fg-towards-unconstrained-human-object-interaction/https://vaquero.me/publications/2026-fg-towards-unconstrained-human-object-interaction/AnyHOI studies human-object interaction detection without a predefined interaction vocabulary, using multimodal LLMs and language-to-graph conversion to recover structured triplets from free-form outputs. (IEEE International Conference on Automatic Face and Gesture Recognition, 2026)Thu, 01 Jan 2026 00:00:00 GMTHuman-Object InteractionMultimodal LLMsFrancesco Tonini, Alessandro Conti, Lorenzo Vaquero, Cigdem Beyan, Elisa Riccihttps://vaquero.me/publications/2026-fg-training-free-semantic-multi-object-tracking/https://vaquero.me/publications/2026-fg-training-free-semantic-multi-object-tracking/TF-SMOT composes frozen detection, segmentation tracking, video-language generation, and LLM disambiguation modules to produce semantic tracking outputs without task-specific training. (IEEE International Conference on Automatic Face and Gesture Recognition, 2026)Thu, 01 Jan 2026 00:00:00 GMTMulti-Object TrackingVision-Language ModelsLaurence Bonat, Francesco Tonini, Elisa Ricci, Lorenzo Vaquerohttps://vaquero.me/publications/2026-fg-zero-shot-temporal-action-localization/https://vaquero.me/publications/2026-fg-zero-shot-temporal-action-localization/TeGu compensates for the lack of supervision in zero-shot temporal action localization by exploiting rich textual cues from large language models, distinguishing foreground from background frames without training. (IEEE International Conference on Automatic Face and Gesture Recognition, 2026)Thu, 01 Jan 2026 00:00:00 GMTTemporal Action LocalizationVision-Language ModelsBenedetta Liberatori, Alessandro Conti, Lorenzo Vaquero, Paolo Rota, Yiming Wang, Elisa Riccihttps://vaquero.me/publications/2025-neurips-convis-bench/https://vaquero.me/publications/2025-neurips-convis-bench/ConViS-Bench evaluates video similarity along semantic concepts, pairing concept-level human scores with free-form descriptions for interpretable video comparison. (Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2025)Wed, 01 Jan 2025 00:00:00 GMTVideo UnderstandingBenchmarksBenedetta Liberatori, Alessandro Conti, Lorenzo Vaquero, Yiming Wang, Elisa Ricci, Paolo Rotahttps://vaquero.me/publications/2025-acmmm-dynamic-scoring-hoi-detection/https://vaquero.me/publications/2025-acmmm-dynamic-scoring-hoi-detection/DYSCO is a training-free HOI detector that combines textual and visual interaction representations in a multimodal registry with dynamic multi-head scoring. (ACM International Conference on Multimedia, 2025)Wed, 01 Jan 2025 00:00:00 GMTHuman-Object InteractionVision-Language ModelsFrancesco Tonini, Lorenzo Vaquero, Alessandro Conti, Cigdem Beyan, Elisa Riccihttps://vaquero.me/publications/2025-ibpria-enhancing-multi-object-tracking-segmentation-masks/https://vaquero.me/publications/2025-ibpria-enhancing-multi-object-tracking-segmentation-masks/A lost-track recovery architecture uses segmentation masks and a transformer-based mask selector to preserve tracks when detectors fail in crowded scenes. (Iberian Conference on Pattern Recognition and Image Analysis, 2025)Wed, 01 Jan 2025 00:00:00 GMTMulti-Object TrackingSegmentationManuel Bendaña, Lorenzo Vaquero, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2025-iccv-superpowering-open-vocabulary-object-detectors-x-ray-vision/https://vaquero.me/publications/2025-iccv-superpowering-open-vocabulary-object-detectors-x-ray-vision/RAXO adapts RGB open-vocabulary detectors to X-ray imagery with training-free visual descriptors and introduces DET-COMPASS for large-scale X-ray OvOD evaluation. (IEEE/CVF International Conference on Computer Vision, 2025)Wed, 01 Jan 2025 00:00:00 GMTOpen-Vocabulary DetectionX-ray VisionPablo Garcia-Fernandez, Lorenzo Vaquero, Mingxuan Liu, Feng Xue, Daniel Cores, Nicu Sebe, Manuel Mucientes, Elisa Riccihttps://vaquero.me/publications/2024-eccv-lost-and-found/https://vaquero.me/publications/2024-eccv-lost-and-found/BUSCA plugs into online tracking-by-detection systems to recover objects missed by detectors through proposal generation and decision-transformer association. (European Conference on Computer Vision, 2024)Mon, 01 Jan 2024 00:00:00 GMTMulti-Object TrackingOnline TrackingLorenzo Vaquero, Yihong Xu, Xavier Alameda-Pineda, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2024-acmmm-al-gtd/https://vaquero.me/publications/2024-acmmm-al-gtd/AL-GTD reduces gaze target annotation cost by combining supervised and self-supervised signals in an active learning acquisition function with pseudo-labeling. (ACM International Conference on Multimedia, 2024)Mon, 01 Jan 2024 00:00:00 GMTGaze Target DetectionActive LearningFrancesco Tonini, Nicola Dall'Asen, Lorenzo Vaquero, Cigdem Beyan, Elisa Riccihttps://vaquero.me/publications/2024-iscas-live-demonstration-sram-dnn-cim/https://vaquero.me/publications/2024-iscas-live-demonstration-sram-dnn-cim/A live mixed-signal compute-in-memory demonstration uses 5-bit signed SRAM weights and PWM-coded images for low-power DNN image recognition. (IEEE International Symposium on Circuits and Systems, 2024)Mon, 01 Jan 2024 00:00:00 GMTCompute-in-MemoryEmbedded AIÓ. Pereira-Rial, D. García-Lesta, Lorenzo Vaquero, P. López, Víctor M. Brea, D. Cabellohttps://vaquero.me/publications/2023-pattern-recognition-real-time-siamese-multiple-object-tracker/https://vaquero.me/publications/2023-pattern-recognition-real-time-siamese-multiple-object-tracker/SiamMOTION tracks dozens of arbitrary objects in real time through feature-pyramid proposals, attention, inertia, pairwise depthwise RPN matching, and multi-object penalization. (Pattern Recognition, 2023)Sun, 01 Jan 2023 00:00:00 GMTMulti-Object TrackingReal-Time VisionLorenzo Vaquero, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2023-tci-depth-estimation-image-restoration-defocused-images/https://vaquero.me/publications/2023-tci-depth-estimation-image-restoration-defocused-images/2HDED:NET jointly estimates depth and restores all-in-focus images from defocused input through a shared encoder and parallel task branches. (IEEE Transactions on Computational Imaging, 2023)Sun, 01 Jan 2023 00:00:00 GMTDepth EstimationImage RestorationSaqib Nazir, Lorenzo Vaquero, Manuel Mucientes, Víctor M. Brea, Daniela Coltuchttps://vaquero.me/publications/2022-icip-2hded-net/https://vaquero.me/publications/2022-icip-2hded-net/2HDED:NET performs depth-from-defocus and all-in-focus image restoration in parallel through a shared encoder and two balanced decoder branches. (IEEE International Conference on Image Processing, 2022)Sat, 01 Jan 2022 00:00:00 GMTDepth EstimationImage DeblurringSaqib Nazir, Lorenzo Vaquero, Manuel Mucientes, Víctor M. Brea, Daniela Coltuchttps://vaquero.me/publications/2022-icpr-fast-multi-object-tracking-feature-pyramid-region-proposal-networks/https://vaquero.me/publications/2022-icpr-fast-multi-object-tracking-feature-pyramid-region-proposal-networks/SiamFAST brings feature-pyramid ROI extraction, pairwise depthwise RPN matching, and multi-object penalization to real-time arbitrary multi-object tracking. (International Conference on Pattern Recognition, 2022)Sat, 01 Jan 2022 00:00:00 GMTMulti-Object TrackingReal-Time VisionLorenzo Vaquero, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2021-pattern-recognition-tracking-more-than-100-arbitrary-objects/https://vaquero.me/publications/2021-pattern-recognition-tracking-more-than-100-arbitrary-objects/SiamMT scales visual object tracking to dozens of arbitrary objects by reusing feature computations and introducing efficient crop-resize and pairwise similarity operators. (Pattern Recognition, 2021)Fri, 01 Jan 2021 00:00:00 GMTMulti-Object TrackingDeep LearningLorenzo Vaquero, Víctor M. Brea, Manuel Mucienteshttps://vaquero.me/publications/2020-icpr-siammt/https://vaquero.me/publications/2020-icpr-siammt/SiamMT is a Siamese convolutional architecture for tracking multiple arbitrary objects in real time by sharing frame features and using pairwise cross-correlation. (International Conference on Pattern Recognition, 2020)Wed, 01 Jan 2020 00:00:00 GMTMulti-Object TrackingSiamese NetworksLorenzo Vaquero, Manuel Mucientes, Víctor M. Breahttps://vaquero.me/publications/2018-wgml-deep-learning-video-object-detection-tracking/https://vaquero.me/publications/2018-wgml-deep-learning-video-object-detection-tracking/Brief summary of deep convolutional approaches for small-object detection, real-time multi-object tracking, and integrated traffic monitoring. (Machine Learning Workshop Galicia, 2018)Mon, 01 Jan 2018 00:00:00 GMTObject DetectionMulti-Object TrackingBrais Bosquet, Mauro Fernández-Sanjurjo, Lorenzo Vaquero, Daniel Cores, Víctor M. Brea, Manuel Mucientes