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

Output results #8

Closed LiAoChUnLiNs closed 2 years ago

LiAoChUnLiNs commented 2 years ago

Does the pink part in the output graph represent an abnormal part?

aseuteurideu commented 2 years ago

Yes

LiAoChUnLiNs commented 2 years ago

There are many contradictory labels in the avenue dataset. The same event is marked as abnormal before and normal later. For example, camera shake, people on the lawn, some frames of packet loss, people dancing, children jumping.

aseuteurideu commented 2 years ago

The label is the original one. But indeed I remember reading one or two papers (but I can't remember exactly which papers exactly for now) mentioning about this label problem and use their own labels instead. But since most papers use the original labels, we use the original labels.

LiAoChUnLiNs commented 2 years ago

thank you.This is really a good idea, especially when there is a speed anomaly.