A5065572304- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Gamma-ray bursts and supernovae
- Multimodal Machine Learning Applications
- Agriculture and Rural Development Research
- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Astrophysics and Cosmic Phenomena
- Generative Adversarial Networks and Image Synthesis
- Image Retrieval and Classification Techniques
- Anomaly Detection Techniques and Applications
- Medical Image Segmentation Techniques
- Water management and technologies
- Pulsars and Gravitational Waves Research
- Context-Aware Activity Recognition Systems
- Face recognition and analysis
- Galaxies: Formation, Evolution, Phenomena
- Neutrino Physics Research
- Speech and dialogue systems
- Nuts composition and effects
- Advanced Vision and Imaging
- Astro and Planetary Science
- COVID-19 diagnosis using AI
- Astrophysical Phenomena and Observations
- Astronomical Observations and Instrumentation
Tencent (China)
2019-2024
Centre National de la Recherche Scientifique
2008-2009
Aix-Marseille Université
2008
National Agricultural Technology Institute
1989-1994
Laboratoire de Recherche Scientifique
1977
Anna Needs Neuroblastoma Answers
1977
Université de Rennes
1974
Large-scale pre-trained models have achieved remarkable success in various computer vision tasks. A standard approach to leverage these is fine-tune all model parameters for downstream tasks, which poses challenges terms of computational and storage costs. Recently, inspired by Natural Language Processing (NLP), parameter-efficient transfer learning has been successfully applied However, most existing techniques primarily focus on single-task adaptation, despite limited research multi-task...
Multi-Task Learning (MTL) is designed to train multiple correlated tasks simultaneously, thereby enhancing the performance of individual tasks. Typically, a multi-task network structure consists shared backbone and task-specific decoders. However, complexity decoders increases with number To tackle this challenge, we integrate decoder-free vision-language model CLIP, which exhibits robust zero-shot generalization capability. Recently, parameter-efficient transfer learning methods have been...
Large-scale pre-trained vision models (PVMs) have shown great potential for adaptability across various downstream tasks. However, with state-of-the-art PVMs growing to billions or even trillions of parameters, the standard full fine-tuning paradigm is becoming unsustainable due high computational and storage demands. In response, researchers are exploring parameter-efficient (PEFT), which seeks exceed performance minimal parameter modifications. This survey provides a comprehensive overview...
We use three years of data from the Supernova Legacy Survey (SNLS) to study general properties core-collapse and type Ia supernovae. This is first such using "rolling search" technique which guarantees well-sampled SNLS light curves good efficiency for supernovae brighter than $i^\prime\sim24$. Using host photometric redshifts, we measure supernova absolute magnitude distribution down luminosities $4.5 {\rm mag}$ fainter normal SNIa. spectroscopy light-curve fitting discriminate against...
Timmes, Brown & Truran found that metallicity variations could theoretically account for a 25% variation in the mass of 56Ni synthesized Type Ia supernovae (SNe Ia), and thus large fraction scatter observed SN luminosities. Higher-metallicity progenitors are more neutron-rich, producing stable burning products relative to radioactive 56Ni. We develop new method estimating bolometric luminosity yield SNe use it test theory with data from Supernova Legacy Survey. find average does drop high...
In real world, training data for person re-identification (Re-ID) is collected discretely with spatial and temporal variations, which requires a model to incrementally learn new knowledge without forgetting old knowledge. This problem called lifelong (LReID). Variations of illumination background images each task exhibit task-specific image style lead task-wise domain gap. addition missing from the tasks, gap key factor catastrophic in LReID, ignored existing approaches LReID. The tends gap,...
<i>Aims. <i/>We present 139 spectra of 124 Type Ia supernovae (SNe Ia) that were observed at the ESO/VLT during first three years Canada-France-Hawaï Telescope (CFHT) supernova legacy survey (SNLS). This homogeneous data set is used to test for redshift evolution SN spectra, and will be in SNLS 3rd year cosmological analyses.<i>Methods. <i/>Spectra have been reduced extracted with a dedicated pipeline uses photometric information from deep CFHT (CFHT-LS) reference images trace, sub-pixel...
Greedy-NMS inherently raises a dilemma, where lower NMS threshold will potentially lead to recall rate and higher introduces more false positives. This problem is severe in pedestrian detection because the instance density varies intensively. However, previous works on don't consider or vaguely factor of existent nearby pedestrians. Thus, we propose \heatmapname (\heatmapnameshort ), which pinpoints objects each proposal with Gaussian distribution, together \nmsname, dynamically eases...
The Contrastive Language-Image Pre-training (CLIP) has recently shown remarkable generalization on "zero-shot" training and applied to many downstream tasks. We explore the adaptation of CLIP achieve a more efficient generalized action recognition method. propose that key lies in explicitly modeling motion cues flowing video frames. To end, we design two-stream block capture spatial information at same time. And then, obtained are utilized drive dynamic prompts learner generate motion-aware...
The Canada-France-Hawaii Telescope Supernova Legacy Survey (SNLS) has created a large homogeneous database of intermediate redshift (0.2 < z 1.0) type Ia supernovae (SNe Ia). SNLS team shown that correlations exist between SN rates, properties, and host galaxy star formation rates. now been combined with photometric catalog an optical cluster to investigate the possible influence clustering on rate, over above expected effect due dependence SFR through morphology-density relation. We...
The evolution of Diffusion Models has dramatically improved image generation quality, making it increasingly difficult to differentiate between real and generated images. This development, while impressive, also raises significant privacy security concerns. In response this, we propose a novel Latent REconstruction error guided feature REfinement method (LaRE^2) for detecting the diffusion-generated We come up with Reconstruction Error (LaRE), first reconstruction-error based in latent space...
Greedy-NMS inherently raises a dilemma, where lower NMS threshold will potentially lead to recall rate and higher introduces more false positives. This problem is severe in pedestrian detection because the instance density varies intensively. However, previous works on don't consider or vaguely factor of existent nearby pedestrians. Thus, we propose Nearby Objects Hallucinator (NOH), which pinpoints objects each proposal with Gaussian distribution, together NOH-NMS, dynamically eases...
Multi-Task Learning (MTL) is designed to train multiple correlated tasks simultaneously, thereby enhancing the performance of individual tasks. Typically, a multi-task network structure consists shared backbone and task-specific decoders. However, complexity decoders increases with number To tackle this challenge, we integrate decoder-free vision-language model CLIP, which exhibits robust zero-shot generalization capability. Recently, parameter-efficient transfer learning methods have been...
Résumé— Les relations entre certaines caractéristiques physiques du trayon et l'aptitude à la traite sont étudiées sur 30 vaches de race Française Frisonne Pie Noire.La cinétique d'écoulement lait pendant est enregistrée séparément pour chaque quartier.Les propriétés élastiques l'ex- trémité mesurées en appliquant des pressions succions les tissus l'aide divers appareils d'une conception simple.Les quartiers arrière ont temps dé- bits plus élevés que l'avant (P < 0,001 ).Cela provient...
Pendant 8 semaines, 48 brebis Lacaune sont utilisées pour étudier l'effet du matériau constituant le manchon (silicone ou caoutchouc) et poids gobelet (plastique acier inox) sur volume de repasse manuelle la fréquence des chutes gobelets.-Le n'affecte pas manuelle.L'accroissement provoque une légère diminution (8,4 p. 100) seulement avec en silicone.Avee notre matériel expérimental, ces deux facteurs ne semblent être d'une grande importance pratique vidange mamelle durant traite.-Quel que...
établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics privés.
— Selon le plan en carré latin 6 x 6, 18 vaches ont été traites pendant périodes d'une semaine avec manchons-trayeurs expérimentaux qui diffèrent entre eux seulement par dia- mètre et l'élasticité du corps.Deux modèles de manchons sont utilisés : petit grand diamètre inté- rieur.Chaque modèle est moulé 3 versions différentes souple, normale (commercialisée) dure des parois 1,3; 2,3 3,3 mm d'épaisseur.La cinétique d'écoulement lait la traite enregistrée automatiquement.Avant après...
The Contrastive Language-Image Pre-training (CLIP) has recently shown remarkable generalization on "zero-shot" training and applied to many downstream tasks. We explore the adaptation of CLIP achieve a more efficient generalized action recognition method. propose that key lies in explicitly modeling motion cues flowing video frames. To end, we design two-stream block capture spatial information at same time. And then, obtained are utilized drive dynamic prompts learner generate motion-aware...
Large-scale pre-trained models have achieved remarkable success in various computer vision tasks. A standard approach to leverage these is fine-tune all model parameters for downstream tasks, which poses challenges terms of computational and storage costs. Recently, inspired by Natural Language Processing (NLP), parameter-efficient transfer learning has been successfully applied However, most existing techniques primarily focus on single-task adaptation, despite limited research multi-task...
Ce travail a pour but la description des mouvements d'un manchon utilisé traite mécanique brebis en fonction de pulsation appliquée.Les matériels d'enregistrements et utilisés laboratoire sont schématisés aux figures i 2 .A chacune 5 vitesses (V) 6 0 , 90 120 jo x8o p /rnn, o enregistrements effectués faisant croître le rapport (R) 20 p. 100 à 8 .L'étude paramètres mesurés (fig. 3et 4 ) montre que :-La considération R V est insuffisante quantifier pulsation.-A chaque combinaison (R, V)...
Currently in object detection, deep learning based detectors are gaining their momentum. However, the supervision involved widely-used anchor paradigm within detection pipeline is inadequate. Traditional opt for densely picking anchors to increase training samples faster convergence and better quality. dense scheme requires extra computational budget which renders it infeasible lightweight detectors. To address problem, inspired by cognitive consistency, we propose a novel Particle Swarm...
La stabilité du vide de traite a été étudiée sur une machine à traire expérimentale dont on fait varier le volume intérieur, V, 80 1 z8o et la capacité, C, pompe vide, 140 N1 /mn à. 86 0 Nl!mn.Le test pratiqué consiste introduire rapidement dans l'installation un d'air 10 NI en branchant pot lait « standard » 20 au niveau lactoduc (fig. ).La perturbation produite est enregistrée par des capteurs pression placés près petit tuyau 2 ).Les résultats présentés figure 3 aux tableaux i montrent...