Extract roads from satellite images

For this problem, we provide a set of satellite/aerial images acquired from GoogleMaps. We also provide ground-truth images where each pixel is labeled as {road, background}. Your goal is to train a classifier to segment roads in these images, i.e. assign a label {road=1, background=0} to each pixel. Please see detailed instructions on the course github.

Dataset

File descriptions

See Resources section:

• training.zip - the training set consisting of images with their ground truth
• test_set_images.zip - the test set
• sampleSubmission.csv - a sample submission file in the correct format
• mask_to_submission.py - script to make a submission file from a binary image
• submission_to_mask.py - script to reconstruct an image from the sample submission file

The sample submission file contains two columns:

• The first column corresponds to the image id followed by the x and y top-left coordinate of the image patch (16x16 pixels)

• The second column is the label assigned to the image patch

Evaluation

Your algorithm is evaluated according to the following criterion:

• F1 score (this combines the two numbers of precision and recall)

Rules

Each participant is allowed to make 5 submissions per day. If you particpate as a team, the whole team gets 5 submissions, not 15 as the rules page states. Failed submissions (e.g. wrong submission file format) do not count.