Closed lsh1994 closed 4 months ago
I used some features from the main branch which aren't in a release yet. You can install using pip install git+https://github.com/microsoft/torchgeo.git
Regarding the conclusion of the paper, I personally don't think so. Many of the models have additional parameters and layers in other ways so nothing is really apples to apples. The main point of the paper is just that nobody appears to be comparing to a good strong baseline of an older architecture. There's not a large jump in performance when using recent proposed architectures.
thanks for you work!I have some questions: