How to train this net about anomalydetectioncvpr2018

[kjsr7]

Extract features by using the C3D model version 1.0. Using the extracted features from fc-6 layer from C3D, run the Save_C3DFeatures_32Segments.m in this repository, that condenses the features of fc6 layer into 32 segments for each video. Then for training, use the Training_Anomaly_detector.py which takes the features of the videos in the Anomaly_train.txt. For understanding the structure of the neural network, you need to look through the research paper mentioned in the README

[VivaaindreanNg]

Extract features by using the C3D model version 1.0. Using the extracted features from fc-6 layer from C3D, run the Save_C3DFeatures_32Segments.m in this repository, that condenses the features of fc6 layer into 32 segments for each video. Then for training, use the Training_Anomaly_detector.py which takes the features of the videos in the Anomaly_train.txt. For understanding the structure of the neural network, you need to look through the research paper mentioned in the README

Hello there,

Just a question. So right before I begin to execute TrainingAnomalyDetector_public.py, I had to extract the features for (any) video by executing C3D-v1.0 and then run the Save_C3DFeatures_32Segments.m, right? Because the extracted C3D features in the SampleVideo folders constitute only a small amount of videos.

[kjsr7]

Yes you are right.

[VivaaindreanNg]

Yes you are right.

Hey there.

I've tried to run the C3D extractor for C3D-v1.0, but for some reason I could not locate the extract_image_features.bin after recompiling the Makefile.config, here's my issue: facebookarchive/C3D#429 .

[arash9900]

follow this link
https://github.com/rajanjitenpatel/C3D_feature_extraction