docs/Data.md

We need more dataset, please refer to the Open-Sora-Dataset for details.

v1.0.0

Text-to-Video

We open source v1.0.0 all the training data, the annotations and the original video can be found here.

These data consist of segmented video clips, with each clip obtained through center cropping. The resolution of each clip is 512×512. There are 64 frames in each clip, and their corresponding captions can be found in the annotation files.

We present additional details in report and Open-Sora-Dataset.

Class-condition

In order to download UCF-101 dataset, you can download the necessary files in here. The code assumes a ucf101 directory with the following structure

UCF-101/
    ApplyEyeMakeup/
        v1.avi
        ...
    ...
    YoYo/
        v1.avi
        ...

Un-condition

We use sky_timelapse, which is an un-condition datasets.

sky_timelapse
├── readme
├── sky_test
├── sky_train
├── test_videofolder.py
└── video_folder.py