Pre-trained model is showing high error when tested on NWPU dataset

[ashwingopinath]

Hi @jia-wan
The paper reports MAE of 79.3 and MSE of 346.1, but on testing with the pretrained model provided, I got MAE 148 and and MSE of 767 on the NWPU val set. Could you let me know if this is expected of the model, or whether the provided pth file is not final.

[jia-wan]

Hi Ashwi,

Thanks for your interest. The model is trained on UCF dataset. That's why it doesn't perform well on NWPU-Crowd. Can you evaluate it on the UCF? If the model is not correct, I will update it. Sorry for the confusion.