Why is my MOCAS accuracy much higher than the accuracy reported in the paper?

[shui27345-jpg]

When using the Husformer algorithm to evaluate the MOCAS dataset, why is the accuracy far greater than that in the paper (for a single subject, after 5 training epochs, the accuracy reaches 99%)? Is this due to a large number of duplicate data in the MOCAS dataset downloaded from the official website? I would like to ask: How was the data processed to calculate the accuracy in the paper? How were duplicate data removed—by deleting duplicate data across all five modalities, or mainly deleting duplicates in a specific key modality? Was the model trained and tested using data from a single subject, or using all non-duplicate data?@zdz0086 @SMARTlab-Purdue @wonsu0513 @R7-Robot

[wiaobang]

Hello, have you tested this Husformer algorithm on the DEAP dataset? The results I got were far worse than the accuracy mentioned in the paper.

[git-boring]

Hello, I'd like to ask how you modified and got the code running. I want to run it on the DEAP dataset.