> Fragments of different folders.

Hey, I would like to have some clarification on the following rule of the challenge:
- Fragments of different folders must not be used when reconstructing a particular disk.

Does this mean that for the training I can not use the fragments of the different tracks ? or that within a track I can not use the other disks when training for a disk ? or this only applies to inference and not training.

¿ Could you express explicitly what can be used for each disk in training and inference ?
thankyou very much.

Posted by: Fgarciaa @ July 3, 2024, 1:04 p.m.

We designed the rule to ensure that each disk is reconstructed in isolation: only the pieces in its disk number for the specific track you are participating in must be considered. You cannot use pieces belonging to other disks to aid the reconstruction of the current disk.
For instance, if you are reconstructing disk 5 of the test set of track 2, you cannot use any file from track 1 or any other file from other disks in track 2 except for disk 5.
Let us know if you have any doubts or questions.

Posted by: nicolo.didomenico @ July 3, 2024, 1:11 p.m.

perfect, thankyou.

So if I understood there are no rules in the training phase, but on inference only the items of a particular disk can be used, right?

Posted by: Fgarciaa @ July 3, 2024, 1:41 p.m.

During training limitations are more relaxed, but during testing you must strictly use only the files you are allowed to use according to the disk you are trying to reconstruct and to the sub-track of your choice.
In practice, if you are trying to reconstruct disk 5 in the test set and your algorithm participates in sub-track 2.1, you can only use the polylines present in track2/test/5.
Conversely, if your algorithm is instead participating in sub-track 1.2, you can only use the raw images present in track1/test/5.
Obviously, you cannot use any part of the test disks during training.

Posted by: nicolo.didomenico @ July 3, 2024, 2:47 p.m.
Post in this thread