search

KaiyangZhou / vsumm-reinforce

AAAI 2018 - Unsupervised video summarization with deep reinforcement learning (Theano)
MIT License
137 stars 36 forks source link

Can I use the generated model to generate a summary of long videos? #13

Open idlerwang opened 6 years ago

idlerwang commented 6 years ago

Hello, I want to train the model with the public tvsum dataset and summe dataset, and send an hour long video frame into the model to generate a video summary. But I did this, I don't know if it's a dataset problem or something else, I found the video summary of the first ten minutes before the model was generated.

Datartic commented 5 years ago

@idlerwang how did you use the model on your own video to generate a summary?

SinDongHwan commented 5 years ago

@syedimad1998 if you want to generate a summary using your video, you should have features, change_points, n_frames, n_frame_per_seg, picks

Datartic commented 5 years ago

@SinDongHwan I am new to this and I don't understand how to extract all this ..do you have any code ?

SinDongHwan commented 5 years ago

@syedimad1998
Features : extract features using CNN. change_points : use KTS using extracted feature, to get change points. n_frames : number of frames of one video. n_frame_per_seg : number of frames each segments. picks : every 15 indicise.

my repository is editing to use custom videos. now i can't work. i will edit ASAP.