overview
play

Overview Yandex Services Car Detection Yandex.Taxi 3D Car - PowerPoint PPT Presentation

Overview Yandex Services Car Detection Yandex.Taxi 3D Car Detection Yandex Technologies Windup SDC Yandex Taxi Upcoming Plans Algorithms Design Evolvement since1989 search engine technology 1997 yandex.ru


  1. Overview › Yandex Services › Car Detection › Yandex.Taxi › 3D Car Detection › Yandex Technologies › Windup › SDC Yandex Taxi › Upcoming Plans › Algorithms Design Evolvement

  2. since1989 search engine technology 1997 yandex.ru world-class information search on-demand transport services navigation products mobile applications

  3. Search Media Geo Services Informational Personal Disk Mail Market Navi Taxi Kinopoisk Afisha Browser Search Health Weather Ad Tech Maps Parking Launcher Zen Music Auto.ru Jobs TV Program Transport Traffic Travels Realty Flights Trains

  4. Yandex Taxi – one of the largest online taxi services in Russia & CIS 2 – 5 min >200 000 6 countries 149 average active cars cities waiting time

  5. Technologies › Machine learning (MatrixNet) › Behaviour analytics technology (Crypta) › Speech recognition › Real-time bidding platform › Language recognition and › Weather forecasting synthesis (SpeechKit) technology (Meteum) › Antivirus technologies › Personal recommendations › Computer vision technology (Disco) › Duplicate image detection

  6. Algorithms Design Evolvement Labelling ML Deployment Data constant ongoing improvement process

  7. Car Detection Pipeline Labelling ML Deployment Data

  8. Data for Car Detection DATA DATA Collection

  9. Labelling for Car Detection Objects Labelling detection

  10. ML for С ar Detection ML Yolo SSD Modifications

  11. Deployment for Car Detection Deployment Drive PX Tensor RT FuseNet

  12. 3D Car Detection Pipeline Lidar Detection Labelling ML Deployment Data Data Recollection

  13. Data for 3D Detection Lidar Detection Lidar Point Cloud

  14. ML for 3D Car Detection Lidar Bird View (BV) Prior boxes classification Concatenation Prior boxes coordinates regression Image (RGB) ROI Pooling

  15. Method Moderate Easy Hard Runtime MV3D (lidar 0,77 0,858 0,689 240ms only) MV3D 0,769 0,86 0,684 360ms Ours (bev only) 0,749 0,821 0, 736 95ms

  16. Deployment for 3D Car Detection Deployment TensorRT Graph freezing DrivePX

  17. Wind-up Data › image segmentation Constant Labelling › improvement depth estimation process ML › end-to-end Deployment

  18. Upcoming Plans more sensors => с ity >=10 cars fleet test-drives growth more experiments 2018

  19. Thank you! Questions? Anton Slesarev Lead computer vision scientist slesarev@yandex-team.ru +7 926 312-89-55

Recommend


More recommend