search

Roc-Ng / DeepMIL

Real-world Anomaly Detection in Surveillance Videos CVPR2018 UCF-Crime dataset
119 stars 11 forks source link

`gt-ucf.npy`为什么不采用帧级的标注，以及片段级标注是如何生成的？ #19

Closed whystopped closed 2 years ago

whystopped commented 2 years ago

UCF-Crime数据集给出了帧级的标注，为什么不直接采用帧级的标注进行比较呢

16帧的片段级标注又是如何生成的呢？是这16帧超过一半为异常，就标注为异常吗？

Roc-Ng commented 2 years ago

用i3d 或者c3d提取特征的时候，non-overlaping的16帧提取一个feature。用repeat方式扩充到16倍即可。

Chase Wu @.***> 于2021年12月21日周二 11:22写道：

UCF-Crime数据集给出了帧级的标注，为什么不直接采用帧级的标注进行比较呢

16帧的片段级标注又是如何生成的呢？是这16帧超过一半为异常，就标注为异常吗？

— Reply to this email directly, view it on GitHub https://github.com/Roc-Ng/DeepMIL/issues/19, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFTZQQFSIPTLBQL65PHI7PDUR7XI7ANCNFSM5KPEZUWA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you are subscribed to this thread.Message ID: @.***>

whystopped commented 2 years ago

用i3d 或者c3d提取特征的时候，non-overlaping的16帧提取一个feature。用repeat方式扩充到16倍即可。

感谢！我之前理解错了

botao-cao commented 2 years ago

你好，可以问一下你有没有在ShanghaiTech数据集上跑过呀，为啥我跑出来AUC有93+

whystopped commented 2 years ago

你好，可以问一下你有没有在ShanghaiTech数据集上跑过呀，为啥我跑出来AUC有93+

这个模型在SHT上跑不到93吧，应该85上下