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.