Skip to content

mvrl/weakly-supervised-segmentation

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 

Weakly Supervused Building Segmentation from Overhead Images

This is the repo for our wekaly supervised building segmentation, accepted at IGARSS 2019.

Dataset

We use the disaster response dataset, released as the mapping challenege:. You should be able to get the data form this website. If you have any trouble in acquiring the dataset, please contact us.

How to use this code

There are several files, comments at the top of each file explain the purpose.

All the settings are stored in config.py. It is expected that you will forst train a model and then you can run visualization code.

To train a model, specify dataset path, training settings (level of supervision, loss function, batch size etc) in config.py. You will also specify a directory in which the trained model will be saved. Once the training finishes, a log file and loss curves will be saved in that folder.

After training, you can run the visualize_trained.py file, which will load the trained model and save some visual results in that folder.

Citation

If you find this paper or code helpful, please cite this paper:

M. Usman Rafique, Nathan Jacobs, "Weakly Supervised Building Segmentation From Aerial Images", In: IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2019.

People

Please feel free to contact us for any question or comment.

M. Usman Rafique

Nathan Jacobs

Permission

The code is provided for academic purposes only without any guarantees.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages