search

aseuteurideu / STEAL

Official implementation of "Synthetic Temporal Anomaly Guided End-to-End Video Anomaly Detection" (ICCV Workshops 2021: RSL-CV).
23 stars 3 forks source link

Dear author #2

Closed LeoVict closed 2 years ago

LeoVict commented 2 years ago

Hello, I used your parameter test and did not get the effect proposed in the paper. Is it because the code has not been updated

aseuteurideu commented 2 years ago

Hmm, it should be the paper's version. Could you give more detail? What is the command you are running? And what is you pytorch version?

LeoVict commented 2 years ago

I just downloaded the trained parameters and ran the "evaluate" code. ped2 AUC is just 86.2%

aseuteurideu commented 2 years ago

Hmmm weird.. Just now I tested on other system. Fresh with new git clone, downloaded the dataset and weight from link in the README. And still got 98.4% AUC. What is your pytorch version? I once saw other repo has problem with AUC number because of the pytorch version.