How is the accuracy of the three models of Zero-shot method in Table 1 evaluated?

[zhangzscn]

How is the accuracy of the three models using Zero-shot method in Table 1 evaluated?
It seems that there is no relevant evaluation method mentioned in the paper.

[lucamodica]

@zhangzscn As far as I understood, the baseline would be the same pipeline of OpenEMMA but witout the CoT reasoning, so you only rely on ego status and from camera image.