I conducted the experiment according to the paper, but there is a gap between the experimental results and the paper. I hope to get your help.

cvlab-yonsei / MNAD

An official implementation of "Learning Memory-guided Normality for Anomaly Detection" (CVPR 2020) in PyTorch.

333 stars 82 forks source link

I conducted the experiment according to the paper, but there is a gap between the experimental results and the paper. I hope to get your help. #38

Open renqi1120 opened 3 years ago

renqi1120 commented 3 years ago

Hello author: I conducted a reconstruction experiment, but there was a gap between the results of the paper. I set parameters as ： method recon;loss_compact 0.01;loss_separate 0.01;t_length 1;alpha 0.7; th 0.015 . Experimental AUC results : ped2 87. 355%; avenue 72.122%; shanghai 68.086% . How can I achieve the expected experimental results. Thank you for your attention.

LiAoChUnLiNs commented 1 year ago

I suspect that the label on the avenue may be incorrect. Many normal labels give abnormal