Project Details

Topografická analýza obrazu s využitím metod hlubokého učení

Project Period: 1. 7. 2019 - 30. 6. 2022

Project Type: grant

Code: LTAIZ19004

Agency: Ministry of Education, Youth and Sports Czech Republic

Program: INTER-EXCELLENCE - Podprogram INTER-ACTION

English title
Deep-Learning Approach to Topographical Image Analysis
Type
grant
Keywords

image geo-localization, topographic information, image registration, deep-learning, computer vision

Abstract

The project focuses on the current problems of computer vision, especially on visual localization in the natural environment. The visual location of the camera in the outdoor environment is not a fixed issue today, although it offers a wide range of attractive applications from automatic image comprehension, to expanded reality applications to navigation of self-governing vehicles and airplanes. The project aims to research new methods for locating cameras based on the multimodal data registration, especially photographic information, synthetic rendered images, depth information and field models using current machine learning methods, especially deep neural networks (DNN). In addition to the use of terrain data in the form of graphical models, an alternative of predictive depth information from an input photograph will be explored. The CPhoto @ FIT Group has been dealing with the long-standing problem and has deep experience in research and application. The Israeli partner also offers unique data sets indispensable for DNN training.

Team members
Čadík Martin, doc. Ing., Ph.D. (UPGM FIT VUT) , research leader
Brejcha Jan, Ing., Ph.D. (UPGM FIT VUT)
Lysek Tomáš, Ing. (UPGM FIT VUT)
Polášek Tomáš, Ing. (UPGM FIT VUT)
Tomešek Jan, Ing. (UPGM FIT VUT)
Publications

2024

  • DEKEL Shay, KELLER Yosi and ČADÍK Martin. Estimating Extreme 3D Image Rotations using Cascaded Attention. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle: IEEE Computer Society, 2024, pp. 2588-2598. ISBN 979-8-3503-5301-3. Detail
  • BOBÁK Petr, ČMOLÍK Ladislav and ČADÍK Martin. Reinforced Labels: Multi-Agent Deep Reinforcement Learning for Point-Feature Label Placement. IEEE Transactions on Visualization and Computer Graphics, vol. 30, no. 9, 2024, pp. 5908-5922. ISSN 1077-2626. Detail

2023

  • POLÁŠEK Tomáš, ČADÍK Martin, KELLER Yosi and BENEŠ Bedřich. Vision UFormer: Long-Range Monocular Absolute Depth Estimation. Computers and Graphics, vol. 111, no. 4, 2023, pp. 180-189. ISSN 0097-8493. Detail

2022

  • TOMEŠEK Jan, ČADÍK Martin and BREJCHA Jan. CrossLocate: Cross-Modal Large-Scale Visual Geo-Localization in Natural Environments using Rendered Modalities. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). Waikoloa: Institute of Electrical and Electronics Engineers, 2022, pp. 2193-2202. ISBN 978-1-6654-0477-8. Detail
  • RAJASEKARAN Suren Deepak, KANG Hao, ČADÍK Martin, GALIN Eric, GUÉRIN Eric, PEYTAVIE Adrien, SLAVÍK Pavel and BENEŠ Bedřich. PTRM: Perceived Terrain Realism Metric. ACM Transactions on Applied Perception, vol. 19, no. 2, 2022, pp. 1-22. ISSN 1544-3558. Detail

2021

  • POLÁŠEK Tomáš, HRŮŠA David, BENEŠ Bedřich and ČADÍK Martin. ICTree: Automatic Perceptual Metrics for Tree Models. ACM Transactions on Graphics (TOG), vol. 40, no. 6, 2021, pp. 1-15. ISSN 0730-0301. Detail
  • AHMAD Touqeer, EMAMI Ebrahim, ČADÍK Martin and BEBIS George. Resource Efficient Mountainous Skyline Extraction using Shallow Learning. In: Proceedings of the International Joint Conference on Neural Networks (IJCNN). Hoffman Estates: Institute of Electrical and Electronics Engineers, 2021, pp. 1-9. ISBN 978-1-6654-3900-8. Detail

2020

  • BREJCHA Jan, LUKÁČ Michal, HOLD-GEOFFROY Yannick, WANG Oliver and ČADÍK Martin. LandscapeAR: Large Scale Outdoor Augmented Reality by Matching Photographs with Terrain Models Using Learned Descriptors. In: Computer Vision - ECCV 2020. Lecture Notes in Computer Science, vol. 12374. Cham: Springer Nature Switzerland AG, 2020, pp. 295-312. ISBN 978-3-030-58525-9. Detail
  • BOBÁK Petr, ČMOLÍK Ladislav and ČADÍK Martin. Temporally Stable Boundary Labeling for Interactive and Non-Interactive Dynamic Scenes. Computers and Graphics, vol. 91, no. 10, 2020, pp. 265-278. ISSN 0097-8493. Detail
Products

2022

  • CrossLocate: Cross-Modal Large-Scale Visual Geo-Localization in Natural Environments using Rendered Modalities, software, 2022
    Authors: Tomešek Jan, Čadík Martin, Brejcha Jan Detail
  • ICTree: Automatic Perceptual Metric for Tree Models, software, 2022
    Authors: Čadík Martin, Polášek Tomáš Detail

2020

  • Immersive Trip Reports, software, 2020
    Authors: Brejcha Jan, Čadík Martin Detail
  • LandscapeAR, software, 2020
    Authors: Brejcha Jan, Čadík Martin Detail
Back to top