pictorial-maps-retinanet

https://github.com/narrat3d/pictorial-maps-retinanet

Science Score: 65.0%

This score indicates how likely this project is to be science-related based on various indicators:

✓
CITATION.cff file
Found CITATION.cff file
✓
codemeta.json file
Found codemeta.json file
✓
.zenodo.json file
Found .zenodo.json file
✓
DOI references
Found 3 DOI reference(s) in README
○
Academic publication links
○
Academic email domains
✓
Institutional organization owner
Organization narrat3d has institutional domain (narrat3d.ethz.ch)
○
JOSS paper metadata
○
Scientific vocabulary similarity
Low similarity (11.0%) to scientific vocabulary

Last synced: 11 months ago · JSON representation ·

Repository

Basic Info

Host: GitHub
Owner: narrat3d
License: mit
Language: Python
Default Branch: master
Homepage: http://narrat3d.ethz.ch/detection-of-pictorial-map-objects-with-cnns/
Size: 93.8 KB

Statistics

Stars: 0
Watchers: 0
Forks: 0
Open Issues: 0
Releases: 0

Created over 2 years ago · Last pushed over 2 years ago

Metadata Files

Readme License Citation

Detection of sailing ships on historic maps with RetinaNet

This is code for the article Detection of Pictorial Map Objects with Convolutional Neural Networks. Visit the project website for more information.

Ships_on_map

Installation

Requires Python 3.6.x
Requires CUDA Toolkit 9.0 and corresponding cuDNN
Download this project
pip install -r requirements.txt
pip install <path to this project>
cd <path to this project>
python setup.py build_ext --inplace

Troubleshooting

Download and install Microsoft Build Tools (see “All Downloads” > "Tools for Visual Studio")

Inference

Download trained model and set SHIPDETECTIONWEIGHTS_PATH to the downloaded model in config.py
Run detect_ships.py <input folder with images of historic map> <output folder for text files and images with detected bounding boxes>

Training

Download training data and set DATA_FOLDER to the downloaded folder in config.py
Download trained coco weights for RetinaNet and set COCOWEIGHTSPATH to the downloaded model in config.py
Adjust LOG_FOLDER in config.py. The trained models will be stored in this folder.
Optionally adjust properties like scales (e.g. SCALEARRAYS = [[2**0, 2**(1/3), 2**(2/3)]]), number of runs (e.g. RUNNRS = ["1st"]), configuration in config.py (CONFIG_KEYS = ["small"])
Run training.py to train the ship detector

Evaluation

Run model_converter.py to convert trained models into inference models
Run evaluation.py to predict bounding boxes and scores for detecting ships (optionally enable the save_image flag to visualize detected and ground truth bounding boxes on the images)
Run coco_metrics.py to calculate COCO metrics

Source

https://github.com/fizyr/keras-retinanet (Apache License, Copyright by Hans Gaiser)

Modifications

Use of higher ResNet blocks in models\resnet.py and higher pyramid levels in utils\anchors.py to detect smaller objects on images
Parametrization of scales, strides, and sizes so that it can be trained in multiples runs with different configurations

Citation

Please cite the following article when using this code: @article{schnuerer2021detection, author = {Raimund Schnürer, René Sieber, Jost Schmid-Lanter, A. Cengiz Öztireli and Lorenz Hurni}, title = {Detection of Pictorial Map Objects with Convolutional Neural Networks}, journal = {The Cartographic Journal}, volume = {58}, number = {1}, pages = {50-68}, year = {2021}, doi = {10.1080/00087041.2020.1738112} }

Owner

Name: narrat3d
Login: narrat3d
Kind: organization
Email: schnuerer@ethz.ch

Website: http://narrat3d.ethz.ch/
Repositories: 1
Profile: https://github.com/narrat3d

Codebase for the doctoral project "Storytelling with Animated Interactive Objects in Real-time 3D Maps"

Citation (CITATION.cff)

cff-version: 1.2.0
message: Please cite the following works when using this code.
preferred-citation:
  authors:
    - family-names: Schnürer
      given-names: Raimund
    - family-names: Sieber
      given-names: René
    - family-names: Schmid-Lanter
      given-names: Jost
    - family-names: Öztireli
      given-names: A. Cengiz
    - family-names: Hurni
      given-names: Lorenz
  doi: 10.1080/00087041.2020.1738112
  identifiers:
    - type: doi
      value: 10.1080/00087041.2020.1738112
    - type: url
      value: https://doi.org/10.1080/00087041.2020.1738112
    - type: other
      value: urn:issn:0008-7041
  title: Detection of Pictorial Map Objects with Convolutional Neural Networks
  url: https://doi.org/10.1080/00087041.2020.1738112
  date-published: 2020-09-11
  year: 2021
  month: 8
  issn: 0008-7041
  issue: '1'
  journal: The Cartographic Journal
  languages:
    - en
  start: '50'
  end: '68'
  type: article
  volume: '58'

GitHub Events

Total

Last Year

Dependencies

requirements.txt pypi

tensorflow ==1.10.0
tensorflow-gpu ==1.10.0

setup.py pypi

Pillow *
cython *
keras *
keras-resnet *
opencv-python *
progressbar2 *
scipy *
six *

ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Open Source Science