Can we use other dataset for training?

[zhaorui0907]

The Youtube dataset is too big.I want to know if I can use other dataset such as UCF101 for training? I have test your trained model on UCF101,but the result is not good.

[yclai1114]

Hi, may i ask how you train on other dataset as i get error when running train.py

[zhaorui0907]

@yclai1114 hi,I didn't train.I don't how to train on other dataset either.But I find that if we want to train,we need a file in dataset/video.jon for our dataset.I don't know how to get it.

[Amysiat]

@zhaorui0907 I would also like to use my own data set to train the network, but do not know how to do it, do you solve this problem?

[albertomontesg]

To train your own datasets it would be necessary to dig into the code and overwrite the functions which read the videos in the first place to read the videos at the location and the format of your own dataset. The important thing to take into account is the dataset information which all the code is adapt to the ActivityNet dataset (in a json file) but for other datasets can be different.

[govzhz]

@albertomontesg I want to train on gestures dataset, but extract the features from the C3D Network need a c3d_sports_weights, Do I need to train a gesture c3d model first? How can I do that? thx. (I use c3d_sports_weights to train can only get 17% acc in validation set..)