When setting the load_tensors parameter to `True`, it is expected that the `self.data_root=PosixPath('/root/autodl-tmp/video-dataset-disney/videos')` contains two folders named `video_latents` and `prompt_embeds`. However, these folders were not found. Please make sure to have prepared your data correctly using `prepare_data.py`.

[DylanDDeng]

System Info / 系統信息

Cuda 12.1, Python 3.12

Information / 问题信息

• [X] The official example scripts / 官方的示例脚本
• [ ] My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

I use official disney dataset and run train_text_to_video_lora.sh, but get the error like " When setting the load_tensors parameter to True, it is expected that the self.data_root=PosixPath('/root/autodl-tmp/video-dataset-disney/videos') contains two folders named video_latents and prompt_embeds. However, these folders were not found. Please make sure to have prepared your data correctly using prepare_data.py. " How can I solve this error? Thanks for help!

Expected behavior / 期待表现

If can help me solve my error, really thanks.

[gloic]

You need to run prepare_dataset.sh before training, it will pre-process the prompts and videos into video_latents and prompt_embeds.

[DylanDDeng]

Thanks for your help! But when I finish running prepare_dataset.sh before training, I find the videos change in the output directory. The original video content is Disney black and white but after doing the pre-process, the video content is like the following file. .

I am not sure if this is right and don't know why the videos changed. Even though they do the embedding work, I think the contents of the videos should be same.

Could you help me to solve this? really thanks !

[sayakpaul]

The content of the serialized videos should not not affected qualitatively. Could this is be some video codec problem? I am able to open the serialized videos on my VSCode.