DSNet
DSNet copied to clipboard
li-plus
Has anyone re-experimented feature extraction from the raw video?
As I have say in #12 , when I re-extract the features to train the network, the f1 score of the model is only about 0.3. Is this normal? Has anyone re-experimented feature extraction from the raw video?
As I have say in #12 , when I re-extract the features to train the network, the f1 score of the model is only about 0.3. Is this normal? Has anyone re-experimented feature extraction from the raw video?
Have you solved this problem? In my other two experiments（https://github.com/StevRamos/video_summarization；https://github.com/e-apostolidis/PGL-SUM）, the effect of using the features extracted by myself is very poor. Even if you train yourself, the effect is not good. Later, it was found that it was not the characteristic problem, but that the same gtscore and change_points could not be obtained. I use gtscore and change_points in h5, as well as the googlenet features extracted by myself, to achieve the effect. But you will find that even if you use the feature of all zeros and gtscore and change_points in h5, you can get better results.
As I have say in #12 , when I re-extract the features to train the network, the f1 score of the model is only about 0.3. Is this normal? Has anyone re-experimented feature extraction from the raw video?
Have you solved this problem? In my other two experiments（https://github.com/StevRamos/video_summarization；https://github.com/e-apostolidis/PGL-SUM）, the effect of using the features extracted by myself is very poor. Even if you train yourself, the effect is not good. Later, it was found that it was not the characteristic problem, but that the same gtscore and change_points could not be obtained. I use gtscore and change_points in h5, as well as the googlenet features extracted by myself, to achieve the effect. But you will find that even if you use the feature of all zeros and gtscore and change_points in h5, you can get better results.
Did you employ any other backbone network?
the effect of using the features extracted by myself is very poor
have you found any solution to this problem ? if you did please let me know. Thanks ,
