search

tianyu0207 / RTFM

Official code for 'Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning' [ICCV 2021]
326 stars 78 forks source link

UCSD Ped2 Train and Test set #70

Open joos2010kj opened 2 years ago

joos2010kj commented 2 years ago

Previous work [16, 79] re-formulate the dataset for weakly supervised anomaly detection by randomly selecting 6 anomaly videos and 4 normal videos into the train set, with the remaining as test set. We report the mean results over 10 times of this process.

Hello!

Can I ask how you randomly selected the videos for the train set, as there is no standard for randomness mentioned? Like, did you use numpy.random.seed(0)? If possible, could you provide the 60 anomaly videos and 40 normal videos used for the process?

tianyu0207 commented 2 years ago

Hi, you can randomly select the videos multiple times (over 10 times) and the performance should be similar regardless of the seed. I didn't keep consistent with the previous paper because they didn't provide the dataset either. So I just do some random selection and compare with them. Although I would say this is a bit unfair.