ChenYi99 / EgoPlan

BSD 3-Clause "New" or "Revised" License
51 stars 6 forks source link

Regarding downloading the Ego4D dataset #4

Closed takenpeanut closed 2 months ago

takenpeanut commented 3 months ago

should I download the full video dataset (namely using --datasets full_scale annotations) or just a subset of it?

ChenYi99 commented 3 months ago

We only use a portion of the videos from the Epic-Kitchens-100 and Ego4D datasets. You can determine which videos are needed for the training and validation sets by referring to the "video_id" field in our data samples. Additionally, please note that we directly use the officially released RGB frames instead of raw videos for Epic-Kitchens.

zulihit commented 3 months ago

Hi, I have the same question? Can I understand it this way that I only need to download the data in P01 and the data in v1_288p?

EPIC-KITCHENS └── P01 └── rgb_frames └── P01_01 ├── frame_0000000001.jpg └── ... Ego4D └──v1_288p ├── 000786a7-3f9d-4fe6-bfb3-045b368f7d44.mp4 └── ...

We only use a portion of the videos from the Epic-Kitchens-100 and Ego4D datasets. You can determine which videos are needed for the training and validation sets by referring to the "video_id" field in our data samples. Additionally, please note that we directly use the officially released RGB frames instead of raw videos for Epic-Kitchens.

zulihit commented 3 months ago

Can you provide more detailed explanations on which part and how to download Ego4D? Thank you very much for your contribution.

ChenYi99 commented 3 months ago

Apologies for the confusion. You need to download the pre-extracted RGB frames data from all the participants in Epic-Kitchens, not just P01. Similarly, you need to download the v1_288p version of videos from Ego4D.

As for which specific video corresponding data you need to download, you can refer to the video_id field of our data samples for download (for instance, PXX_YY is the style of video_id in Epic-Kitchens, where XX represents the participant number and YY the video number, and 000786a7-3f9d-4fe6-bfb3-045b368f7d44 is the style of video_id in Ego4D).

ChenYi99 commented 3 months ago

You can follow this page to download the Ego4D dataset. The more detailed document can be found here.

After signing Ego4D license and downloading the CLI. You can run the following command to download the full-scale videos of version 1:

ego4d --output_directory="~/ego4d_data" --datasets full_scale --metadata --version v1

Additionally, you can specify video UIDs using the "--video_uids" flag to reduce the download size.

takenpeanut commented 2 months ago

Thanks for the clarification!