crowdai/mapping-challenge-starter-kit

Starter kit for getting started in the crowdAI Mapping Challenge

mapping-challenge-starter-kit's Issues

Add instructions for round 2

Hi,
@spMohanty it would be nice to have instructions for round 2.
Cheers

Submission failure

Hello, I just make three submissions (5795, 5796 and 5697).

However, the server post the following error:
Argument 'bb' has incorrect type (expected numpy.ndarray, got list).

Actually, JSON does not support "numpy.ndarray" format, right? I don't know why it posts this kind of error as my first try (submission 5731, only contains very few annotations) has no errors. I see the admin @spMohanty also had this kind of error before (submission 4617). Does anyone know what results in this problem and how to solve this?

Thank you very much!

About training data specification (metadata)

What is the specification of the training data (e.g. spatial resolution or the satellite that took the images)?
Also, the training data seems like it is blue shifted compared to other dataset that I have and the pretrained model has poor performance in that dataset. Can you provide the data specification so that I can modify my dataset to be closer to this one. Thanks! :)

Add stats about the dataset

Like total number of buildings, total area covered, etc.

CrowdAIExecuteFunctionError: Argument 'bb' has incorrect type (expected numpy.ndarray, got list)

What is bb ?
CrowdAIExecuteFunctionError: Argument 'bb' has incorrect type (expected numpy.ndarray, got list)
submit .json format
numpy.array is not serialized

Thanks
Peter

Generating MS COCO like annotations for the dataset

I would like ask that what method was followed in generating the MS COCO like annotations for the satellite images used in the challenge.
If possible, please provide me the code or the approach because I am working on a similar project to segment objects like cars, ships, etc. from satellite images and all I have is the bounding boxes of those objects and I want to convert them into MS COCO format for training with Mask RCNN.

Thanks in advance