some questions regarding the auxiliary dataset:
1. you state that the background dataset does not contain any targets. so how is it possible that segment 123 (in your example) from the training set appears as 1000123 in the background dataset? how did you remove the target
from the segment? same goes for synthetic - if it is only based on highSNR then LowSNR segments cannot appear there.
2. regarding the synthetic set you state that "the distribution of this set is identical to the distribution of the High SNR segments in the training set", but the set also contains experiments data which has a different distribution. so can i assume this sentance only referes to the equivalent segments?
Erez.Posted by: erez.sharon @ July 29, 2020, 6:21 a.m.
1) The background is only similar, not exactly the same. Background segments are recorded using the same sensor and at the same time as the matching training segments, but at a slightly shifted location where the tracked target obviously doesn't exist.
2) Synthetic - To add more Low SNR segments we created synthetically Low SNR segments from the High SNR segments that you have in the "Training Set" and in the "Auxiliary Experiment Set". We added noise to the High SNR segments to simulate Low SNR segments.
You are correct about the distribution, the note is referring only to the part that is from the "Training set", The other part is the same as the "Auxiliary Experiment Set" High SNR segments (which contains only humans).
(MAFAT Challenge Team)