In this competition, we will see how well your facade segmentation network can perform. You are encouraged to try and improve your kaggle result, but you are not allowed to use the kaggle ground truth data as part of training.

Submitting works using the DICE metric and uses a run length encoding (RLE) output. Basically, if you were to produce a pixelwise mask for all 5 classes, it would lead to really large text files filled with values like:

[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]

[1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1]

[1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1]

[1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 1]

[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]

To avoid this, we are providing you a function to encode the above masks as a run length encoding. The run length encoding turns a list of 10 1's into something like [1, 10]. In this case 1 represents starting at the first index and 10 means 'running' for the next 10 instances. 

Because you are able to use our function for run length encoding, this should not be an issue. But when you look at the sample submission, don't be worried by how strange it looks.

To create a submission, you need to generate test output for the 'kaggle' dataset we have provided  for you in the starter code. See the 'Evaluation' page for more information on how it will be scored.