Discussion

Dear organizers,
 

Please tell us whether the following two points are possible due to the competition rules.

• Learning using external data other than speedplus
  • For example, is it permissible to use a training dataset that we add to speedplus from our own Tango 3D model?
• Labeling of lightbox / sunlight and use in training
  • For example, is it permissible to label a lightbox image with a teacher label and use it as part of a training dataset?

Thanks a lot in advance!

Best regards.

Hello aihara,

we would like our participants to compete fairly and on even ground. Spending some time and effort to collect additional external data or making creative use of resources outside the challenge is encouraged as long as such resources are generally available to all competitors. However, if you have for example - by coincidence - a replica of the exact same satellite-model that we use exclusively available to you, it seems an unfair advantage. A similar reasoning would apply to a digital model, unless you build one yourself specifically for the competition out of the provided dataset (i.e. not by using closed external sources).

Realistically speaking, we have no way to enforce this rule though other than appeal to your sportsmanship. An interesting and fair option would also be to make exclusive resources available to everyone by sharing them in this discussion board.

Of course, sharing also open resources that could be useful to anyone is encouraged.

Hello marcusm,

Thank you for your answer. 
I understand that it is important for all participants to have fair and equal competition.

Currently, I don't have my own Tango 3D model yet. I also don't have my own labeled dataset, but I wanted to know about the possibilities of the rules.

I am grateful for the very polite answer.
I hope it will be a good competition.

Is it possible to get the 3D model used to generate the dataset? I have built a simulation tool to generate datasets for space objects as part of my master and would like to test if that helps in that challenge. As you already mentioned above, it would not be fair if others do not have access to the same resources. So, would it be possible to make the model public for all teams?

Regards,

Yanick
team hslu_space

Hello Yanick,

The competition was designed with the goal of learning from pixels only so the official model that we used for dataset generation will not be made available.

Thanks for the quick response!

Ok i understand. Is it possible to get the model after the challenge is complete? I am working on a research project and that would help a lot.
On a similar note, will the ground-truth data be released afterwards?

Regards,
Yanick

So far we have not decided about the post-competition phase.