video datasets (UCF101, Kinetics400) download error (downloading nothing)

[sejongpyo]

I'm trying to get some video datasets. I followed the documentation but keep getting an error.

`from torchvision.datasets import UCF101
from torchvision.datasets import Kinetics400
ucf_data_dir = "./data"
ucf_label_dir = "./ucfTrainTestlist"

ucf = UCF101(root = ucf_data_dir, annotation_path = ucf_label_dir, frames_per_clip = frames_per_clip,
step_between_clips = step_between_clips)
kin = Kinetics400(root = ucf_data_dir, frames_per_clip = frames_per_clip,
step_between_clips = step_between_clips)
`

How to fix this?

cc @pmeier

[pmeier]

Hey @sejongpyo. What exactly do you mean by

I'm trying to get some video datasets.

? It seems your iteration does nothing which might be a result of no data available in ucf_data_dir and ucf_label_dir. How are these directories structured?

[sejongpyo]

Hey @sejongpyo. What exactly do you mean by

I'm trying to get some video datasets.

? It seems your iteration does nothing which might be a result of no data available in ucf_data_dir and ucf_label_dir. How are these directories structured?

ucf_data_dir is just a empty folder. Isn't the root path meaning that i'm going to save the dataset? And, ucf_label_dir has ucfTrainTestlist which consist of .txt files of UCF101 dataset.

[pmeier]

Isn't the root path meaning that i'm going to save the dataset?

Nope. torchvision.datasets provides wrappers around datasets so they can be easily used from PyTorch. Most datasets have download=True flag in case you don't already have the files locally. Unfortunately neither UCF101 nor Kinetics400 support download as of now. For Kinectis #3680 will probably land soon and add exactly this.

You can download the UCF101 dataset here. If you want to see this functionality from within torchvision, we are happy to accept a PR adding this.

[sejongpyo]

Unfortunately neither UCF101 nor Kinetics400 support download as of now. For Kinectis #3680 will probably land soon and add exactly this.

You can download the UCF101 dataset here.

Thank you for your help!