Inquiry Regarding Reproducing Results

[suminRoh]

I recently read your paper with great interest and have been thoroughly impressed by your work. To better understand and validate your findings, I attempted to reproduce the results using the code and default setup provided in your paper. However, after running the experiments on the Crema-D dataset, I obtained the following results:
Audio accuracy: 59.54%
Visual accuracy: 75.67%
Fusion accuracy: 77.82%
These results differ from the reported performances in the paper. I followed the provided commands and default setup as outlined but was unable to achieve the same performance metrics you reported.
Could you kindly provide any additional insights or recommendations to help me achieve the performance detailed in your paper?