Feature/GT/Split files for custom dataset

wanboyang / anomaly_detection_LAD2000

25 stars 2 forks source link

Feature/GT/Split files for custom dataset #2

Closed abhishekaich27 closed 2 years ago

abhishekaich27 commented 3 years ago

Hi,

Thank you for providing the extracted features. I was wondering if I can create the same format files for a new dataset: UCSD-Ped1.

From your repo, I can create I3D features of UCSD-Ped1 using this link: https://github.com/wanboyang/anomly_feature.pytorch

Can you help me create the GT files (_frame_label_IET.pickle, video_labelIET.pickle), _trainsplit.txt, and _testsplit.txt for UCSD-Ped1 dataset?

wanboyang commented 2 years ago

UCSD-Ped1 and Ped2 datasets are too small to be divided into fixed training and test sets，following work[1], k-fold cross-validation is recommended. Thanks for your attention. [1] Jia-Xing Zhong, Nannan Li, Weijie Kong, Shan Liu, Thomas H. Li, Ge Li; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 1237-1246