Building Detection MaskRCNN Versions Save

Building detection from the SpaceNet dataset by using Mask RCNN.

v2.0

3 years ago

This release adds:

  • The code for transforming RGB-PanSharpen (16-bit TIF file) into RGB format (8-bit PNG file).
  • SpaceNet_train.py was uploaded instead of former Train.ipynb.
  • Average Precision (AP) was increased to meanAP: 0.940809965993059.
  • The loss was approximate 0.50.
  • A new pre-trained h5 file was released.
  • New outputs were attached. So you can compare them with the former release.

Thanks to those who are relevant to this project, who left a star, and who created an issue!

NOTE:

To attain the new h5 file, you need to send an email to me. If you need help to understand MaskRCNN or Building-Detection-MaskRCNN, do not hesitate to create an issue or to send an email to me.