swathikirans / violence-recognition-avss Goto Github PK

As is mentioned in the paper,

In the proposed model, we used the AlexNet model pre-trained on the ImageNet database as the CNN model for extracting frame level features.

where can we find the use of pre-trained alexnet in the code?

Thank you.

There is no mean to test on your pretrained network if dataset configuration is not specific.

are there links to the datasets that could be placed either in PyTorch or lua repo?

hi , first can you explain why you did 3 times training for your model isn't the goal is one model that can do good in all datasets coz we have one problem space is a violent problem
also how you can say for sure your models is not overfitted and you just reported the test result isn't the best way to show us the training vs test loss also the samples number so small and you make 100 epochs this for sure will lead to overfitting

Hi, How to extract frames from videos as is on paper? based on average. in movie dataset selected 10 frames? how do that?

How to actually use the pre-trained models to detect violence. What should I use?