search

yuz1wan / video_distillation

Official implementation of Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement.
16 stars 2 forks source link

How to train a model with full-dataset and coreset selection strategies? #4

Closed Zi-YuanYang closed 1 month ago

Zi-YuanYang commented 1 month ago

Thanks for the code sharing. This work is interesting, but I met some problems in using the code.

I wander that how to train the model with the full-dataset or with a certain coreset selection strategy? Besides, is the "UCF101actions.pkl" necessary?

yuz1wan commented 1 month ago

Hi! Thanks for your interest in our work!

For full-dataset training, you can use the dataloaders in distill_utils/dataset.py and evaluate_synset function(with mode = 'none') in utils.py.

For coreset selection strategy, we refer to k-center baseline for k-center strategy and herding baseline for herding strategy. We can provide the code we used, but it has not yet been thoroughly organized. We will expedite the organization process and add it to the repository as soon as possible. Should you encounter any issues, please feel free to reach out to me, and I will be more than happy to assist you.

Zi-YuanYang commented 1 month ago

Thanks for your quick response. I'm downloading the Kine-400 dataset, I noticed that this dataset is really large. I wander that how much storage space do I need to store the preprocessed data. I'm afraid that I don't have enough space to store the preprocessed data to continue my reasearch.

yuz1wan commented 1 month ago

The complete Kinetics400 dataset is over 400+G in size, and after preprocessing, it is about 18G.

Zi-YuanYang commented 1 month ago

The complete Kinetics400 dataset is over 400+G in size, and after preprocessing, it is about 18G.

Thanks for the kind information. It really helps a lot. Have a nice day~