Training data

I’m just wondering if you have to use only the training data provided, or can you expand or even exclude the data? I would still want to use the test set to verify that it works but it seems somewhat limiting. That may very well be the point to the exercise though, so I could be out of line here.


I am also wondering if external data is allowed. The rules seem to allow it, as they are similar to the last Amazon Prime Air competition, but I would like to be sure:

Ya this is what I figured that I’d do, use a backbone, an opensource structure, two large training datasets plus the one they provide, and once completed test it with the test data and then have instructions on what I did to reproduce the results upon submission. However, in the case this isn’t allowed I’ve been mulling over ways to accomplish this without anyone else’s datasets or backbones, just a structure that anyone could make. Fingers crossed that it’s allowed because I think it’d be cool to go hardcore with this challenge!


the usage of external data is perfectly allowed and welcomed, provided that the solution remains fully reproductible. The answer posted for the previous Airborne Object Tracking Challenge applies to this challenge as well, including the 3 examples of situations where using external data wouldn’t compromise the reproducibility.

1 Like

Oh that’s awesome thank you so much!! I’ll make sure to provide everything I’m doing in order for it to be reproduced and or verified that I’m not going against the wishes of the challenge.