search

nguyetn89 / Anomaly_detection_ICCV2019

Anomaly Detection in Video Sequence with Appearance-Motion Correspondence
BSD 2-Clause "Simplified" License
53 stars 14 forks source link

Question about W_I and W_F #2

Closed sdjsngs closed 3 years ago

sdjsngs commented 4 years ago

Hi , i am re-implement your paper in pytorch and has some question about W_i and W_f , in section 3.5 of your paper , the weight is obtained on the training data?
For Avenue dataset, when training is done (i.e. 15 epochs), you test the 16 training clips on G model through the 15 epoch's checkpoint to get the out_appe and out_flow , then to get the weight_I and weight_F in these 16 clips on average?
thank you !!!

nguyetn89 commented 4 years ago

Hi, yes! On that way, each weight indicates the impact of its partial score in the combination. The weight is defined as the inverse of the (average) score under the perspective that the lower score should give the higher impact (because the training data contain only frames of normality), and vice versa.