This repository is for Weakly Supervised Video Anomaly Detection via Center-Guided Discriminative Learning(ICME 2020). The original paper can be found (https://ieeexplore.ieee.org/document/9102722) or (https://arxiv.org/abs/2104.07268)
Firstly, thank you for sharing your source code for the community.
I have a question about the parameters to reproduce the results in the experiment.
I trained your code with the default param values on the code, but the ano_false_alarm_all_video (variable all_ano_false_alarm) is different from the 0.10 reported on the paper. Do I need to use any specific parameters?
I have downloaded the i3d features and extracted the features in a local path. I have also changed the "dataset_path" to my local path. But the i3d features still cannot be found. I notice that there is just one file named "dataset" in the extracted file. Here is the error information:
Traceback (most recent call last):
File "/data1/haoyue/codes/remote/Anomaly-Detection/Anomaly_AR_Net_ICME_2020/main.py", line 29, in
train_dataset = dataset(args=args, train=True)
File "/data1/haoyue/codes/remote/Anomaly-Detection/Anomaly_AR_Net_ICME_2020/video_dataset_anomaly_balance_uni_sample.py", line 38, in init
self.videoname = os.listdir(self.feature_path)
NotADirectoryError: [Errno 20] Not a directory: './dataset/dataset/shanghaitech/features_video/i3d/combine'
I got all_auc and far, however, I'm confused about AUC of abnormal video.
Can you explain "AUC_Score_abnormal_video" metric?
I couldn't find it in the paper.