Are you over 18 and want to see adult content?
More Annotations
![A complete backup of loisirs-foret.com](https://www.archivebay.com/archive2/3334fe55-df86-4eb9-9399-7286a30629a4.png)
A complete backup of loisirs-foret.com
Are you over 18 and want to see adult content?
![A complete backup of webzoon.blogspot.com](https://www.archivebay.com/archive2/cb541b3f-a8d7-4c80-a01b-892bb8cd39f7.png)
A complete backup of webzoon.blogspot.com
Are you over 18 and want to see adult content?
![A complete backup of cazarepensiunicostinesti.ro](https://www.archivebay.com/archive2/7081a105-27ef-4f98-a8dc-705dc4f02f88.png)
A complete backup of cazarepensiunicostinesti.ro
Are you over 18 and want to see adult content?
![A complete backup of k-12leadership.org](https://www.archivebay.com/archive2/f82c0f80-6a1b-40a9-92dd-d815fec00507.png)
A complete backup of k-12leadership.org
Are you over 18 and want to see adult content?
Favourite Annotations
![A complete backup of dictatethegame.com](https://www.archivebay.com/archive2/17535a91-d0fa-4631-b8d8-f2ebebe926f4.png)
A complete backup of dictatethegame.com
Are you over 18 and want to see adult content?
![A complete backup of dialcomics.blogspot.com](https://www.archivebay.com/archive2/1133dddd-8317-4f39-8f25-14afb61caf27.png)
A complete backup of dialcomics.blogspot.com
Are you over 18 and want to see adult content?
![A complete backup of montessoriguide.org](https://www.archivebay.com/archive2/9d363649-5a80-4908-a5aa-867e26ef8002.png)
A complete backup of montessoriguide.org
Are you over 18 and want to see adult content?
![A complete backup of friendstrackcalltaxi.com](https://www.archivebay.com/archive2/bd653bfc-a30a-45f1-9289-7b482d22bb2d.png)
A complete backup of friendstrackcalltaxi.com
Are you over 18 and want to see adult content?
![A complete backup of hausarzt-teutschenthal.de](https://www.archivebay.com/archive2/3cab9362-7409-49aa-8c29-e997a7c15302.png)
A complete backup of hausarzt-teutschenthal.de
Are you over 18 and want to see adult content?
![A complete backup of deer5001.blogspot.com](https://www.archivebay.com/archive2/6b6d70a8-3557-45e1-b5db-8f1c35798497.png)
A complete backup of deer5001.blogspot.com
Are you over 18 and want to see adult content?
Text
NAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders.NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION IEEE ICRA 2021 (hybrid format). 3 accepted papers on robotic navigation, multi-finger grasping. Details in news article. WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV 5 th – 9 th January 2021 (Tuesday – Saturday) WACV is a premier international computer vision event. WACV 2021: All sessions will be virtual. Accepted papers: Unsupervised meta-domain adaptation for fashion retrieval, Vivek Sharma, Naila Murray, Diane Larlus, M. Saquib Sarfraz, Rainer Stiefelhagen, Gabriela Csurka Khedari; StacMR: Scene-text aware cross-modal retrieval, Andres Mafla FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. ADAPTIVE THRESHOLDING FOR THE DIGITALDESK PIERRE D. WELLNER July, 1993 Adaptive Thresholding for the DigitalDesk Pierre D. Wellner wellner@europarc.xerox.com Introduction to the problem The image produced by a video camera pointing at black printing on a white VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &NAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders.NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION IEEE ICRA 2021 (hybrid format). 3 accepted papers on robotic navigation, multi-finger grasping. Details in news article. WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV 5 th – 9 th January 2021 (Tuesday – Saturday) WACV is a premier international computer vision event. WACV 2021: All sessions will be virtual. Accepted papers: Unsupervised meta-domain adaptation for fashion retrieval, Vivek Sharma, Naila Murray, Diane Larlus, M. Saquib Sarfraz, Rainer Stiefelhagen, Gabriela Csurka Khedari; StacMR: Scene-text aware cross-modal retrieval, Andres Mafla FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. ADAPTIVE THRESHOLDING FOR THE DIGITALDESK PIERRE D. WELLNER July, 1993 Adaptive Thresholding for the DigitalDesk Pierre D. Wellner wellner@europarc.xerox.com Introduction to the problem The image produced by a video camera pointing at black printing on a white VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &NAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders.NAVER AT EMNLP 2020
5pm – 5:30pm CET (UTC/GMT+01:00) NLP research and openings at NAVER LABS Europe. Laurent Besacier, LIG and NLP group lead at NAVER LABS Europe: NLP research and openings at NAVER LABS Europe. This presentation is intended for potential academic collaborators, PhD or internship candidates. DETECTING LOW-RANK REGIONS IN OMNIDIRECTIONAL IMAGES Omnidirectional Computer Vision 2nd Workshop in conjunction CVPR 2021 publication. Authors: Zoltan Kato, Gabor Nagy, Martin Humenberger andGabriela Csurka
RELEASING FIRST OF A KIND LARGE-SCALE LOCALIZATION Visual localization algorithms estimate the camera pose of a given image which is an important task in robotics, autonomous driving and in augmented reality (AR) and the results of modern visual localization methods have been promising enough for some real-world applications to have been developed such as AR navigation.. Challenges. Indoor environments differ from outdoor scenes in a RISK-SENSITIVE ROBOT NAVIGATION Risk-sensitive robot navigation. We want robots to move around more safely in everyday environments like homes and shops, without harming humans, other robots, their surroundings, or themselves. Simultaneously, we explore how effectively a single policy learned by reinforcement learning can modulate robot behaviour, from risk-averse(cautious
LEARNING VISUAL REPRESENTATIONS WITH CAPTION ANNOTATIONS News: Our paper has been accepted at the European Conference on Computer Vision (ECCV) 2020, 23-28 August 2020: Learning Visual Representations with Caption Annotations. We have released the weights of the ICMLM models pretrained on MS-COCO. Our demo, where you can visualize the attention maps of our ICMLM model for a set of MS-COCO(image
FORCE ENABLES EXTREME PRUNING OF ARTIFICIAL NEURAL FORCE, our new approach to pruning ANNs, implements iterative pruning at initialization. By removing a small number of weights at each step, using the FORCE objective and the gradient approximation, our approach achieves extreme sparsity in the network with a much better sparsity/accuracy trade-off than previous methods. NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &DIANE LARLUS
I obtained a Master degree in Image, Vision and Robotics from UJF/INP (Grenoble, France) in 2005. From 2005 to 2008, I worked as a doctoral candidate in the LEAR group (now named THOTH) at INRIA Grenoble.In 2007, I interned at the JRL/AIST robotics lab (Tsukuba, Japan). I obtained my PhD in 2008 from INP Grenoble.From 2008 to 2010, I worked as a post-doc at TU Darmstadt (Germany).ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. COMPUTER VISION RESEARCH Our research combines skills in machine learning, pattern recognition and computer vision, and we work on multi-disciplinary problems with teams specialised in natural language processing, user experience, ethnography, design and more. Our research efforts may be either long-term in focus, or may tackle problems with concrete and immediate FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
3D Road Layout. We provide the 3D layout of the road surface that we extract from aerial photographs. It contains information regarding the types and precise 3D locations of visual structures on the road surface that are essential for self-driving vehicles, such as lanes, road markings, crosswalks, crossroads and speed bumps.NAVER AT EMNLP 2020
5pm – 5:30pm CET (UTC/GMT+01:00) NLP research and openings at NAVER LABS Europe. Laurent Besacier, LIG and NLP group lead at NAVER LABS Europe: NLP research and openings at NAVER LABS Europe. This presentation is intended for potential academic collaborators, PhD or internship candidates. THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
GLOBAL AI R&D BELT
Global AI R&D Belt. NAVER Labs Europe is a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. By collaborating with different partners we aim to make AI technology in South East Asia and Europe more competitive. We have a strong R&D network built on excellence, common research interests ARTIFICIAL INTELLIGENCE NAVER LABS Europe is the biggest industrial research lab in artificial intelligence in France and a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. NAVER LABS Europe is part of NAVER LABS, creating future technology at NAVER, Korea’s leading internet company and a global innovator.NAVER LABS DATASET
Your browser does not support HTML5 video. NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X & 3D VISION - NAVER LABS EUROPE The research focus of the 3D Vision team lies on the design of methods which combine geometry and learning-based approaches to solve specific real-world challenges such as visual localization, camera pose estimation and 3D reconstruction. Examples for our target applications are robot navigation, indoor mapping, augmented reality (AR) and, more BLOG | NAVER LABS EUROPE NAVER LABS AI research in Europe – a podcast with lab manager Matthias Gallé. Podcast and transcript of Matthias Gallé, head of the NAVER research LAB in Europe who tells us what kind of research is going on in the labs in France and what it’s like to work there. 2021. 27 February 2021. PRAIRIE AND MIAI INTERNATIONAL SUMMER SCHOOL 2020 This is the 3 rd edition of the PAISS summer school. The first one was in 2018, co-organised by Inria and NAVER LABS Europe and held in Grenoble. The second version was held in Paris, in October 2019, and organized by Inria and the institutes PRAIRIE and MIAI.. NAVER LABS is a founding member of two of France’s four 3IA institutes (artificial intelligence interdisciplinary institutesNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
3D Road Layout. We provide the 3D layout of the road surface that we extract from aerial photographs. It contains information regarding the types and precise 3D locations of visual structures on the road surface that are essential for self-driving vehicles, such as lanes, road markings, crosswalks, crossroads and speed bumps.NAVER AT EMNLP 2020
5pm – 5:30pm CET (UTC/GMT+01:00) NLP research and openings at NAVER LABS Europe. Laurent Besacier, LIG and NLP group lead at NAVER LABS Europe: NLP research and openings at NAVER LABS Europe. This presentation is intended for potential academic collaborators, PhD or internship candidates. THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
GLOBAL AI R&D BELT
Global AI R&D Belt. NAVER Labs Europe is a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. By collaborating with different partners we aim to make AI technology in South East Asia and Europe more competitive. We have a strong R&D network built on excellence, common research interests ARTIFICIAL INTELLIGENCE NAVER LABS Europe is the biggest industrial research lab in artificial intelligence in France and a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. NAVER LABS Europe is part of NAVER LABS, creating future technology at NAVER, Korea’s leading internet company and a global innovator.NAVER LABS DATASET
Your browser does not support HTML5 video. NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X & 3D VISION - NAVER LABS EUROPE The research focus of the 3D Vision team lies on the design of methods which combine geometry and learning-based approaches to solve specific real-world challenges such as visual localization, camera pose estimation and 3D reconstruction. Examples for our target applications are robot navigation, indoor mapping, augmented reality (AR) and, more BLOG | NAVER LABS EUROPE NAVER LABS AI research in Europe – a podcast with lab manager Matthias Gallé. Podcast and transcript of Matthias Gallé, head of the NAVER research LAB in Europe who tells us what kind of research is going on in the labs in France and what it’s like to work there. 2021. 27 February 2021. ADVERSARIAL TRANSFER OF POSE ESTIMATION REGRESSION Publication: TASK-CV) workshop, at the European Conference on Computer Vision (ECCV), Glasgow, UK (virtual event), 23 August, 2020NAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
3D Road Layout. We provide the 3D layout of the road surface that we extract from aerial photographs. It contains information regarding the types and precise 3D locations of visual structures on the road surface that are essential for self-driving vehicles, such as lanes, road markings, crosswalks, crossroads and speed bumps. THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
NAVER AT EMNLP 2020
5pm – 5:30pm CET (UTC/GMT+01:00) NLP research and openings at NAVER LABS Europe. Laurent Besacier, LIG and NLP group lead at NAVER LABS Europe: NLP research and openings at NAVER LABS Europe. This presentation is intended for potential academic collaborators, PhD or internship candidates.GLOBAL AI R&D BELT
Global AI R&D Belt. NAVER Labs Europe is a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. By collaborating with different partners we aim to make AI technology in South East Asia and Europe more competitive. We have a strong R&D network built on excellence, common research interests ARTIFICIAL INTELLIGENCE NAVER LABS Europe is the biggest industrial research lab in artificial intelligence in France and a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. NAVER LABS Europe is part of NAVER LABS, creating future technology at NAVER, Korea’s leading internet company and a global innovator.NAVER LABS DATASET
Your browser does not support HTML5 video. NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X & 3D VISION - NAVER LABS EUROPE The research focus of the 3D Vision team lies on the design of methods which combine geometry and learning-based approaches to solve specific real-world challenges such as visual localization, camera pose estimation and 3D reconstruction. Examples for our target applications are robot navigation, indoor mapping, augmented reality (AR) and, more BLOG | NAVER LABS EUROPE NAVER LABS AI research in Europe – a podcast with lab manager Matthias Gallé. Podcast and transcript of Matthias Gallé, head of the NAVER research LAB in Europe who tells us what kind of research is going on in the labs in France and what it’s like to work there. 2021. 27 February 2021. ADVERSARIAL TRANSFER OF POSE ESTIMATION REGRESSION Publication: TASK-CV) workshop, at the European Conference on Computer Vision (ECCV), Glasgow, UK (virtual event), 23 August, 2020NAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
NAVER LABS R1 is a mobile mapping system (MMS) designed to create a high definition (HD) map for self-driving vehicles. As a comparatively lightweight MMS, R1 features a variety of sensors including multiple cameras, 2D & 3D LiDAR's, GPS, IMU, FOG and wheel encoders. FROM HANDCRAFTED TO DEEP LOCAL FEATURES FOR COMPUTER An article we recently published on arXiv gives an overview of the evolution of local features – from handcrafted to deep-learning-based methods – and a discussion of several benchmarks and papers that evaluate local features. We also provide references to most of the relevant literature and, whenever possible, link to code and data that are available to the community. PROXY VIRTUAL WORLDS The experimental conclusions are identical to the ones of our CVPR 2016 paper. In fact, the average gap in MOTA for DPMCF is even smaller now (81.0 on real KITTI, 81.2 on VKITTI 1.3.1 clones). 10 Aug. 2016: Update to scene ground truth (v.1.2.1). Small bug fix on poles and transparent shaders impacting only few pixels of the scene groundtruth
TRANSFORMER-BASED META-IMITATION LEARNING FOR ROBOTIC First, we introduce the use of transformer-based sequence-to-sequence policy networks trained from limited sets of demonstrations. Then, we propose to meta-train our model from a set of training demonstrations by leveraging optimization-based meta-learning. Finally, we evaluate our approach and report encouraging results using the recently PROXY VIRTUAL WORLDS VKITTI 2 Virtual KITTI 2 Dataset. Virtual KITTI 2 is a more photo-realistic and better-featured version of the original virtual KITTI dataset. It exploits recent improvements of the Unity game engine and provides new data such as stereo images or scene flow. Blog article: AnnouncingVirtual KITTI 2.
NAVER AT EMNLP 2020
Meet us live at EMNLP. Enter the virtual venue space and meet us on the Gather.town platform (need to be an EMNLP registered attendee and login). Check out the NAVER booth on Rocket chat. Tuesday, 17 th November 2020. 10am-11am UTC + 9:00 (2am-3am CET) Kang Min Yoo,Research Scientist.
THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
VIRTUAL WORLDS AS PROXY FOR MULTI-OBJECT TRACKING ANALYSIS Virtual Worlds as Proxy for Multi-Object Tracking Analysis Adrien Gaidon1 Qiao Wang2 Yohann Cabon1 Eleonora Vig1y 1Computer Vision group, Xerox Research Center Europe, France 2School of Electrical, Computer, and Energy Engineering and School of Arts, Media, and Engineering, Arizona State University, USA fadrien.gaidon,yohann.cabong@xrce.xerox.com qiao.wang@asu.edueleonora.vig@dlr.de
NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X &ALEXANDRE BERARD
Alexandre Berard is on Naver Labs Europe. Join Naver Labs Europe to view Alexandre Berard\'s profileNAVER LABS DATASET
3D Road Layout. We provide the 3D layout of the road surface that we extract from aerial photographs. It contains information regarding the types and precise 3D locations of visual structures on the road surface that are essential for self-driving vehicles, such as lanes, road markings, crosswalks, crossroads and speed bumps. THE REVOLUTION IN PROTEIN STRUCTURE PREDICTION Andrei Lupas is an evolutionary biologist and currently Director of the department of protein evolution at the Max-Planck-Institute forDevelopmental
NAVER AT EMNLP 2020
5pm – 5:30pm CET (UTC/GMT+01:00) NLP research and openings at NAVER LABS Europe. Laurent Besacier, LIG and NLP group lead at NAVER LABS Europe: NLP research and openings at NAVER LABS Europe. This presentation is intended for potential academic collaborators, PhD or internship candidates.GLOBAL AI R&D BELT
Global AI R&D Belt. NAVER Labs Europe is a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. By collaborating with different partners we aim to make AI technology in South East Asia and Europe more competitive. We have a strong R&D network built on excellence, common research interests ARTIFICIAL INTELLIGENCE NAVER LABS Europe is the biggest industrial research lab in artificial intelligence in France and a hub of NAVER’s global AI R&D Belt, a network of centres of excellence in Korea, Japan, Vietnam & Europe. NAVER LABS Europe is part of NAVER LABS, creating future technology at NAVER, Korea’s leading internet company and a global innovator.NAVER LABS DATASET
Your browser does not support HTML5 video. NAVER LABS EUROPE HOMEPAGE NAVER LABS releases world's biggest visual localization dataset of indoor spaces with over 130K images. Dataset built with NAVER LABS mapping robots M1X & 3D VISION - NAVER LABS EUROPE The research focus of the 3D Vision team lies on the design of methods which combine geometry and learning-based approaches to solve specific real-world challenges such as visual localization, camera pose estimation and 3D reconstruction. Examples for our target applications are robot navigation, indoor mapping, augmented reality (AR) and, more BLOG | NAVER LABS EUROPE NAVER LABS AI research in Europe – a podcast with lab manager Matthias Gallé. Podcast and transcript of Matthias Gallé, head of the NAVER research LAB in Europe who tells us what kind of research is going on in the labs in France and what it’s like to work there. 2021. 27 February 2021. ADVERSARIAL TRANSFER OF POSE ESTIMATION REGRESSION Publication: TASK-CV) workshop, at the European Conference on Computer Vision (ECCV), Glasgow, UK (virtual event), 23 August, 2020*
LAVER LABS
* Story
* Projects
* LOG
* Publications
* Career
* Contact
* English
메뉴닫기
ARC
Multi-Robot Intelligence SystemPlay video
A-CITY
Autonomous Everywhere, Everything, EverydayPlay video
AROUND
Autonomous Service Robot PlatformPlay video
M1
Indoor 3D/HD Mapping RobotPlay video
BRAINLESS
ROBOT
Uploading the Robot’s Brain to the CloudPlay video
HYBRID
HD MAPPING
New Solution for Autonomous VehiclesPlay video
0번째 영상 재생1번째 영상 재생2번째 영상 재생3번째 영상 재생4번째 영상 재생5번째 영상 재생CREATING
NEW CONNECTIONS
BY ADVANCING TECHNOLOGY 네이버랩스는 네이버의 미래 기술을 책임지는 R&D 전문 자회사입니다. 한국과 유럽의 우수한 연구자들이 함께 AI, 로보틱스, 자율주행, 3D/HD 매핑, AR 등의 연구에 매진하고 있습니다. 가장 독창적이며 앞 기술력을 통해 사람, 머신, 공간, 정보를 새롭게 연결할 네이버 플랫폼의 미래를 준비하는 것이 우리의 미션입니다. SELF-MOTIVATIED TEAM PLAYER 경계없이 협력하고 스스로 결정하며, 함께 성장 인재를 기다립니다.CAREER
FEATURED NEWS
*
ARC, 멀티 로봇 인텔리전스 시스템 2020.11.25 네이버랩스 네이버랩스에서 고도화하고 있는 멀티 로봇 인텔리전스 시스템 ARC를 소개합니다. Background : 클라우드와 5G라는 기회 그간 네이버의 로봇 연구는 3가지로 집약됩니다. 사람과 로봇 간의 자연스러운 인터랙션, 로봇을 위한 인공지능, 그리고 서비스 로봇의 대중화입니다. 특히 ‘대중화’는 지금 동시대 로봇 연구자들의 공통된 목표일 것입니다. 네이버랩스도 처음부터 이에 집중하여 준비해왔습니다. 로봇을 위한 지도를 만드는 매핑로봇 M1, 맵클라우드와 심층강화학습 기반으로 레이저스캐너없이도 자연스럽게 자율주행할 수 있는 AROUND 플랫폼, 실내에서 별도 인프라 없이 정밀하게 위치 인식을 할 수 있는 Visual Localization 기술 등이 모두 로봇 대중화를 위한 중장기 로드맵하에 진행된 것입니다. 특히 클라우드는 로봇 서비스 대중화의 키가 되는 기술입니다. 사실 클라우드와 로봇을 연결하려는 아이디어는 이미 오래전부터 있어왔지만, 5G 네트워크의 등장은 확실한 전환점이 됩니다. 네이버랩스는 5G를 활용하는 로봇의 잠재력을 가장 먼저 실증했습니다. 5G 초저지연 통신 성능 활용을 극대화했을 때 로봇의 운동과 제어까지 클라우드가 대신할 수 있을 거라 예측했고, 2019년 CES에서 세계 최초의 5G 브레인리스 로봇 데모로 이를 증명했습니다. 이 기술이 로봇 서비스의 양상을 바꿀 기폭제가 될 것이라 확신했고, 곧바로 상용화 시스템 개발에 착수했습니다. 그 결과가 바로 ARC입니다. 모든 로봇들을 동시에 똑똑하게 해주는 인텔리전스 시스템 AI, Robot, Cloud의 약자인 ARC에는 로봇과 사람의 공존을 위해 필요한 최신 알고리즘과 고정밀 데이터가 담겨있습니다. 초저지연 네트워크를 통해 ARC에 연결되는 것만으로 수많은 로봇들이 동시에 똑똑해질 수 있습니다. GPS가 통하지 않는 실내에서도 현재 위치를 정확하게 알려주고, 이동, 측위, 태스크 수행을 위한 계획과 처리를 네이버 클라우드 플랫폼에 올라간 ARC가 대신합니다. 또한 공간 및 서비스 인프라와도 실시간으로 연동되어, 로봇과 사람 사이의 모든 온-오프라인 환경을 이해하 관리합니다. 우리는 ARC가 로봇들을 위한 NAVER 서비스와 같다고 생각합니다. 마치 사람들이 네이버의 검색, 지식iN, 지도 등에 접속해 필요한 정보를 얻는 것처럼, 로봇들은 클라우드에 띄워진 ARC를 통해 사람과 공존하고 공간을 이해하기 위한 정보를 실시간으로 얻을 수 있게 됩니다. 또한 ARC는 브레인리스 로봇 기술을 통해 로봇의 운동과 제어를 관장하는 소뇌 역할까지 클라우드가 대신할 수 있도록 하기 때문에, 고가의 센서나 컴퓨팅 파워에 들어가는 로봇 각각의 제작비를 낮추면서도 퍼포먼스를 끌어올릴 수 있겠죠. 특히 미니치타처럼 소형화, 경량화가 중요한 로봇도 거대한 데이터센터가 두뇌 역할을 대신하여 더 똑똑해질 수 있습니다. 이 시스템을 좀 더 자세히 들여다보면, 모든 로봇들의 눈이 되어 줄 ARC eye와 모든 로봇들의 두뇌가 될 ARC brain으로 구성되어 있습니다. 로봇들의 천리안이 될 ARC eye 로봇들이 서비스를 제공하기 위해서는 잘 움직일 수 있는 모빌리티 능력이 필수입니다. 뛰어난 자율주행 기술과 함께, 지금 어디에 있는지를 정확하게 인식하고 어디로 가야하는지 효율적으로 경로를 계획할 수 있어야 합니다. ARC eye는 로봇들의 천리안(telegnosis)입니다. 로봇들이 어디에 있더라도 현재 위치를 정확하게 인식해 가장 효율적인 동선을 로봇에게 전달할 수 있게 합니다. GPS가 통하지 않는 실내에서도 문제없습니다. 우리가 간판이나 주변의 모습을 보 지금 서있는 곳이 어딘지 파악하듯이, 카메라 이미지만으로 로봇의 정확한 위치를 파악할 수 있는 Visual Localization이라는 AI 기술 때문입니다. 로봇들이 공유하는 두뇌, ARC brain ARC brain은 모든 로봇들의 두뇌가 되어 이동, 측위, 태스크 수행을 위한 계획과 처리를 대신합니다. 또한 빌딩 인프라 및 서비스 서버와 실시간으로 연결되어 로봇들이 가장 효율적이며 자연스럽게 사용자들에게 서비스를 제공하는 역할도 하게 됩니다. 이렇게 로봇 자체에서 수행하던 기능을 ARC 시스템으로 분리할 수 있게 되면 흥미로운 장점이 생깁니다. 로봇이 서비스를 수행하기 위해 필요한 기능, 즉 어플리케이션을 클라우드를 통해 원격으로 쉽게 업데이트하거나 전환할 수 있다는 것입니다. 예를 들어, 아침에 택배 배달을 하던 로봇이 점심에는 카페에서 일할 수 있도록 전환해 운영하는 것도 ARC brain으로 쉽게 가능해집니다. 또한 이런 식으로 점점 더 많은 기능을 ARC가 대신하게 되면 로봇 하드웨어 자체는 점차 더 심플한 구성으로 제작될 수 있게 됩니다. ARC를 고도화하는 4개의 요소 우리가 이 새로운 멀티 로봇 인텔리전스 시스템의 고도화에 큰 자신감을 갖는 이유는 4가지입니다. 첫 번째는 네이버랩스가 다년간 축적해온 기술과 데이터입니다. 우리는 고정밀 데이터, 로봇을 위한 인공지능, HRI, 하드웨어와 소프트웨어, 서버 엔지니어링에 이르는 모든 영역에서 내재화된 독자 기술을 구축해왔습니다. 두 번째는 네이버 클라우드 플랫폼입니다. 아마 네이버의 앞선 클라우드 컴퓨팅 기술과 인프라가 없었다면 네이버의 로봇 서비스 대중화를 위한 도전은 시작되기도 어려웠을 것입니다. 향후 준공 예정인 하이퍼스케일 데이터센터 각 세종이 10년 뒤에는 전국 모 로봇들의 브레인센터가 되어있을 것이라 기대하 있습니다. 세 번째는 네이버 제2사옥입니다. ARC의 첫번째 버전인 ARC-1이 네이버 제2사옥에서 첫 상용화됩니다. 아무리 기술이 뛰어나더라도 단기간의 파일럿 테스트만으로는 다양한 환경과 상황에 대처하기가 어렵습니다. 세계 최초로 로봇 친화형 빌딩으로 준공 중인 네이버 제2사옥은 ARC 시스템을 가장 빠르게 고도화하고 완성도를 높일 수 있는 최고의 테스트필드입니다. 이 곳은 미래의 일상 공간이 로봇 서비스의 대중화에 맞춰 어떤 준비를 해야할지 보여주는 레퍼런스가 될 것입니다. 그리 마지막으로 네이버의 다양한 서비스들입니다. 결국 서비스 로봇의 핵심은, 종국에 서비스일 수밖에 없습니다. 유저에게 실질적 혜택과 새로운 경험을 제공해야 합니다. 앞으로 로봇은 마치 스마트폰이 그랬던 것처럼, 다양한 네이버 서비스를 연결하는 허브가 될 것입니다. 멀티 로봇 인텔리전스 시스템 ARC를 통해 AI와 로봇, 클라우드의 새로운 연결이 시작됩니다. ARC 소개페이지 >*
글로벌기계기술포럼에서 사람의 운동지능을 학습하는 앰비덱스와 딜리버리 로봇 어라운드D 공개 2020.11.11 네이버랩스 네이버랩스 석상옥 대표는 ‘2020 글로벌 기계기 포럼 (한국기계연구원 주최)’에서 사람의 운동지능을 학습하기 위한 로봇팔 앰비덱스(AMBIDEX)의 태스크러닝 프로젝트와 새로운 딜리버리 로봇 어라운드D(AROUND D)를 공개했습니다. ‘사람을 위한 기계, 로봇’이란 주제의 이번 포럼을 통해 사람과 로봇이 공존하기 위해 연구해온 네이버랩스의 성과들을 공유했습니다. 특히 발표의 하이라이트는 계속 진화 중인 로봇 앰비덱스입니다. 한국기술교육대학교와 산학협력으로 개발한 앰비덱스는 허리부가 추가되어 작업 반경이 더욱 확장되었고, 센서헤드로 대상을 인식할 수 있으며, 파지 방법을 다양하게 바꿀 수 있는 로봇손인 'BLT 그리퍼'도 장착되었습니다. 외견보다 큰 변화는 사람의 Physical Intelligence를 학습하는 능력입니다. Physical Intelligence란 네이버랩스의 테크컨설턴트이자 MIT 생체 모방 로봇 연구소를 이끌고 있는 김상배 교수가 제시한 개념으로, 그 실행과정을 언어화/수치화하기 어렵지만 사람들이 자연스럽게 행할 수 있는 운동지능을 의미합니다. 빵에 잼을 바르거나, 주머니에서 동전을 꺼내는 것과 같은 일은 사람 스스로 의식하지 않고도 쉽게 행할 수 있는 동작이지만 그 과정과 원리를 언어로 설명하거나 프로그래밍하기는 어렵습니다. 네이버랩스는 이를 위해 사람의 힘 조절 능력을 학습 데이터로 추출 수 있는 앰비덱스 전용 햅틱 디바이스를 개발했습니다. 앰비덱스 태스크티칭 & 러닝 데모 영상 앰비덱스 Perception 연동 기술 데모 이 햅틱 디바이스는 사람의 크기와 동일한 스케일, 사람의 팔과 같은 7자유도 (한쪽 팔 기준), 사람과 로봇 양방향으로 힘이 전달되는 원격 제어 (Bilateral Teleoperation) 등의 특징을 가집니다. 이를 통해 사람이 직접 수행한 데모에서 세심한 힘 조절 데이터를 가져와 로봇의 학습 레퍼런스로 활용할 수 있게 되었습니다. 햅틱 디바이스을 활용해 강화학습 등 다양한 방법으로 로봇을 학습시키는 방식은, 단 하나의 데모를 통해서도 사람의 개입 없이 로봇 스스로 의도에 맞는 작업 수행에 성공할 수 있을 정도로 정확도와 효율이 뛰어납니다. 앰비덱스 태스크러닝 프로젝트 인터뷰 영상 한편 석상옥 대표는 2017년 첫 공개한 네이버의 자율주행 로봇 어라운드 (AROUND) 시리즈의 4번째 모델도 공개했습니다. 딜리버리에 특화된 어라운드D는 고가의 LiDAR 센서 없이도 Vision 기술과 강화학습을 기반으로 자연스러운 자율주행이 가능하며, 이동 및 서비스 과정에서 Gaze를 통해 사람과 직관적인 인터랙션을 할 수 있습니다. 어라운드제로(AROUND-0)라는 HW/SW 기본 플랫폼에 다양한 서비스 어플리케이션을 추가할 수 있도록 설계해 확장성있는 개발이 가능합니다. 또한 네이버 클라우드가 로봇의 두뇌 역할을 대신하는 브레인리스 로봇 기술이 적용된 모델이기도 합니다. 현재 로봇 친화형 빌딩으로 건축중인 네이버 제2사옥에서 사용될 로봇들 중 하나가 될 것이라 밝혔습니다. 이번 포럼을 통해 석상옥 대표는 "네이버의 로봇 연구는 실내/실외/도로와 같이 로봇들이 활동할 공간들의 고정밀 지도 연구로 시작해, 클라우드 기반의 로봇 제어 시스템, 그리 사람의 운동지능을 학습하는 태스크러닝 도구 개발에 이르렀다"며 "언택트 시대에 더욱 조명받 있는 로봇이 일상 공간에 들어오기 위해서는 로봇을 둘러싼 환경, 그리고 공존의 대상인 사람에 대한 연구가 더 필요하다"고 강조했습니다. 2020 글로벌 기계기술 포럼 바로가기>*
서울시 전역 도로 정보, 2,092KM 로드 레이아웃 공개 2020.08.06 네이버랩스 네이버랩스에서 제작한 서울시 전역 4차선 이상 도로 2,092km의 로드 레이아웃입니다. 이 로드 레이아웃은 앞서 공개한 서울시 3D 모델링에서 추출한 것입니다. 아주 정밀한 차선 구조와 노면 기호 정보를 담 있어, 도로 단위가 아닌 차선 단위의 길 안내와 도로 정보 제공까지 가능합니다. 이처럼 대도시 규모의 정밀한 로드 레이아웃을 제작한 것은 큰 의미가 있습니다. 앞으로의 도로/도시 정보화를 위해 반드시 필요한 기술입니다. 자율주행 시대를 앞당기기 위해서도 중요하지만, 도로 위의 안전하면서도 편리한 정보와 서비스 제공에도 다양하게 활용될 수 있습니다. 딥러닝과 비전 (vision) 기술을 통한 정확도 향상 네이버랩스는 작년말 1차로 서울시 전역의 로드 레이아웃 제작을 완료했고, 이번에 공개한 버전은 정확도를 더욱 향상시킨 것으로 오차는 약 16cm 이내입니다. 도로 데이터 추출과 정확도 개선을 위해 다양한 딥러닝과 비전 기술들이 적용되었습니다. 먼저 도로 위의 자동차나 나무 등 불필요한 장애물을 제거해 TrueOrtho(정사영상)를 생성/융합하는 기술이 필요합니다. 여기에 높이를 정제하는 알고리즘을 적용하여 더욱 정확한 도로 정보 데이터를 확보하고, 최종적으로 3D 확장 기술을 통해 로드 레이아웃 지도를 완성합니다. 차선 및 노면 기호 추출에는 AI를 통한 자동화 알고리즘을 별도로 개발하여 적용했습니다. 이를 통해 정확도를 높이는 한편, 제작 시간 및 비용을 절감하는 효과를 얻을 수 있었습니다. 관련 글 : 딥러닝 기술을 통한 항공 영상의 도로 레이아웃 정보 추출 1) 항공영상 융합 기술 : 가려진 노면표지를 식별할 수 있는 TrueOrtho(진정사영상)를 생성하는 기술 2) 도로 영역 DEM 정제 기술 : 도로 위 자동차, 나무 등 불필요한 장애물 제거 및 높이 평탄화 기술 3) Road Layout 2D -> 3D 확장 기술 : 2D 로드 레이아웃 데이터에 도로 높이 정보를 입력하여 3D로 변환하는 기술 4) 노면기호 분류 자동화 기술 : 딥러닝 알고리즘을 적용하여 박스로 표시된 도로 노면 기호들을 자동으로 분류하는 기술 하이브리드 HD 매핑과 3가지 결과물 네이버랩스는 HD맵 제작을 위한 독자적 기술을 가지고 있는데, 항공사진과 MMS 데이터를 융합하는 방식의 하이브리드 HD 매핑 기술입니다. 이 과정에서 항공사진에서 추출한 3D 모델링, 차선과 노면기호를 추출한 로드 레아아웃, 그리고 MMS 데이터까지 결합한 HD맵까지 3가지 형태의 결과물을 제작할 수 있습니다. 각각의 데이터는 디지털 트윈, 정밀한 길 안내, 자율주행 등 여러 목적으로 활용될 수 있습니다. 이미 공개한 서울시 3D 모델링은 버추얼 서울 플랫폼에 적용되어 도시계획심의나 도시바람길 시뮬레이션, IOT센서 소방 시설물 관리 등에 활용하 있습니다. 또한, 로드 레이아웃은 lane-level navigation이나 AR HUD, ADAS 등에 활용될 수 있습니다. 관련 글 : AHEAD (3D AR HUD) demonstration in a real urban environment 네이버랩스는 작년에 이어 올해 이 로드 레이아웃을 활용해 제작한 HD맵 데이터셋 일부를 무상으로 공개 배포한 바 있습니다. 또한 최근 성남시와 AI·자율주행 산업 발전을 위한 업무 협약을 맺고 하이브리드 HD 매핑 기술을 통한 3가지 데이터를 활용해 도심 내 완전자율주행기술 개발과 연구 커뮤니티 확산을 위한 지속적인 협력을 이어나갈예정입니다.
See all
VIDEOS
*
로봇을 만난 서버 엔지니어 - 플랫폼 엔지니어링 팀인터뷰
2020.07.17
*
항공사진 기반 서울시 3차원 모델링 - 3D CITY MODELS OFSEOUL
2020.06.18
*
LOG | VL & AR
2020.05.14
See all
* Youtube
*
* 개인정보 처리방침 * 기업윤리 센터*
사이트 바로가기* NAVER
* NAVER LABS Europe
* © NAVER LABS Corp. * 사이트 바로가기* NAVER
* NAVER LABS Europe
Details
Copyright © 2024 ArchiveBay.com. All rights reserved. Terms of Use | Privacy Policy | DMCA | 2021 | Feedback | Advertising | RSS 2.0