Hello,
The dataset for the Final Round can be found here: https://zenodo.org/record/7269120#.Y2Bk4XXMIqU
The dataset contains 1,000 MNIST networks, and the distribution of attack specifications is the same as in the Primary Round datasets for the Detection Track. The main difference is that the method for inserting trojans is the same method used by one of the winning teams of the Evasive Trojans track.
As specified on the competition website, there is no training set. This is because we hope to evaluate existing detectors developed during the Primary Round, and it also allows us to gauge the impact of distribution shift on method performance.
All the best,
Mantas (TDC co-organizer)