1. ABOUT THE DATASET -------------------- Title: Deep learning for the estimation of water-levels using river cameras: networks and datasets Creator(s): Remy Vandaele[1], Sarah L. Dance[1,2], Varun Ojha[3] Organisation(s): 1. Department of Meteorology, University of Reading, U.K 2. Department of Mathematics, University of Reading, U.K 3. Department of Computer Sciences, University of Reading, U.K Rights-holder(s): University of Reading and Farson Digital Limited Publication Year: 2021 Description: This dataset contains: - The networks weights (weights.zip) that were obtained and used in our papers [1, 2]. We refer to these papers for the methodology used to obtain those weights. Those weights can be used for the binary water semantic segmentation of new images, or for comparison with our methods. - The river camera images (DIGLIS.zip/EVESHAM.zip/STRENSHAM.zip/TEWKESBURY.zip) for the experiments that we presented in [2]. The images can be used for water segmentation and flood analysis purposes. Cite as: Remy Vandaele, Sarah L. Dance, Varun Ojha (2020): Deep learning for the estimation of water-levels using river cameras: networks and datasets. University of Reading. Dataset. http://dx.doi.org/10.17864/1947.282 Related publications: [1] Remy Vandaele, Sarah L. Dance, Varun Ojha; Automated water segmentation and river level detection on camera images using transfer learning; 2020; Proceedings of the DAGM German Conference on Pattern Recognition (Accepted) [2] Remy Vandaele, Sarah L. Dance, Varun Ojha; Deep learning for the estimation of water-levels using river cameras; 2020; Hydrology and Earth System Science (in preparation) Contact: r.a.vandaele@reading.ac.uk, s.l.dance@reading.ac.uk 2. TERMS OF USE ----------------- River camera images (DIGLIS.zip/EVESHAM.zip/STRENSHAM.zip/TEWKESBURY.zip): Copyright 2020 Farson Digital Limited. These images may only be downloaded for personal non-commercial use: https://www.farsondigitalwatercams.com/copyright. Networks weights data (weights.zip): Copyright 2020 University of Reading. These data are licensed by the rights-holder under a Creative Commons Attribution 4.0 International Licence: https://creativecommons.org/licenses/by/4.0/. 3. PROJECT AND FUNDING INFORMATION ---------------------------------- Title: Data Assimilation for REsilient city Dates: 1 Sept 2016 – 31 Jan 2022 Funding organisation: EPSRC Grant no.: EPSRC EP/P002331/1 4. CONTENTS ------------ The "weights.zip" archive contains two repositories: - The deeplabv3 repository contains the networks weights that can be loaded onto a pre-existing DeeplabV3 network architecture. Please refer to the instructions.txt file in the repository for further information about its usage. - The resnet50upernet repository contains the networks weights that can be loaded onto a pre-existing Resnet50-Upernet network architecture. Please refer to the instructions.txt file in the repository for further information about its usage. The DIGLIS/EVESHAM/STRENSHAM/TEWKESBURY.zip archives contain the png images that can be used for water-level segmentation, each named according to the camera location. The naming conventions of the images can allow to retrieve the timestamp of the image at which it was taken: [DATASET]_[YEAR]_[MONTH]_[DAY]_[HOUR].png 5. METHODS -------------------------- We refer to our papers [1, 2] for details of how the networks and images were obtained and used.