https://github.com/akihironitta/xview-pytorch

PyTorch data utilities for object detection with xView Detection Challenge dataset [http://xviewdataset.org/]
https://github.com/akihironitta/xview-pytorch

deep-learning machine-learning pytorch remote-sensing satellite-imagery

Last synced: about 1 month ago
JSON representation

PyTorch data utilities for object detection with xView Detection Challenge dataset [http://xviewdataset.org/]

• Host: GitHub
• URL: https://github.com/akihironitta/xview-pytorch
• Owner: akihironitta
• Created: 2020-02-21T05:10:01.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-22T17:22:58.000Z (almost 5 years ago)
• Last Synced: 2024-12-05T17:52:07.733Z (about 2 months ago)
• Topics: deep-learning, machine-learning, pytorch, remote-sensing, satellite-imagery
• Language: Python
• Homepage:
• Size: 25.7 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.org

Awesome Lists containing this project

README

        
* DESCRIPTION

  This is a PyTorch implementation of data utilities for [[http://xviewdataset.org/][xView Detection Challenge dataset]].

  =XviewDataset= is a subclass of =torchvision.datasets.vision.VisionDataset= similarly to =torchvision.datasets.VOCDetection= and =torchvision.datasets.CocoDetection=.

  

  Given an =index=, =XviewDataset= returns =PIL.Image= and a list of dictionaries in the image.

  Each dictionary describes an object consisting of elements with the keys, ="image_id"=, ="bbox"=, and ="category_id"=.

  | key           | value                                                                   |

  |---------------+-------------------------------------------------------------------------|

  | "image_id"    | int (e.g. =1468=)                                                       |

  | "bbox"        | list of int in =[x, y, width, height]= format (e.g. =[0, 0, 100, 100]=) |

  | "category_id" | int (e.g. =77= for "Building")                                          |

  

* Usage

  #+begin_src sh

  git clone https://github.com/akihironitta/xView-PyTorch.git

  cd xView-PyTorch

  #+end_src

  Just use =XviewDataset= class in your PyTorch code:

  #+begin_src python

  from datasets import XviewDataset

  ds = XviewDataset(root="/path/to/image_dir/",

                    annFile="/path/to/annotation_file.json")

  img, target = ds[0]  # take the 0th image and objects in it.

  print(img)           # 

  print(target)        # [{'image_id': 1468, 'bbox': [2479, 1522, 166, 168], 'category_id': 73}, ...]

  #+end_src

  To view an image with bouding boxes:

  #+begin_src sh

  python view.py 0 --output /path/to/output_dir/

  #+end_src

  

  To save an image with bouding boxes:

  #+begin_src sh

  python view.py 0 --output /path/to/output_dir/

  #+end_src

  You can also view or save multiple images by:

  #+begin_src sh

  python view.py 0 1 2 3

  python view.py 0 1 2 3 --output /path/to/output_dir/

  #+end_src

* SEE ALSO

  The official TensorFlow implementation by DIUx is available at:

  1. [[https://github.com/DIUx-xView/data_utilities][data utilities]]

  2. [[https://github.com/DIUx-xView/xview2018-baseline][baseline models]]