I've evaluated on the social_test split using the given ttm-best.pth. EvalAI gives me {"test_split": {"mAP": 0.5217902698544443, "Acc": 0.5286449623532675}}. This is lower than mAP:53.88% ACC:54.33% as shown in the instructions. I'm curious if this gap is normal, or I am using the wrong checkpoint to evaluate. Any suggestion is greatly appreciated.
Hi! Thank you for the detailed instructions!
I've evaluated on the social_test split using the given ttm-best.pth. EvalAI gives me {"test_split": {"mAP": 0.5217902698544443, "Acc": 0.5286449623532675}}. This is lower than mAP:53.88% ACC:54.33% as shown in the instructions. I'm curious if this gap is normal, or I am using the wrong checkpoint to evaluate. Any suggestion is greatly appreciated.
Thanks again!
Best, Felicia