search

Soldelli / MAD

MAD: A Scalable Dataset for Language Grounding in Videos from Movie Audio Descriptions
MIT License
147 stars 3 forks source link

Where can I download MAD-v2 #14

Open zzz-zq opened 8 months ago

zzz-zq commented 8 months ago

AutoAD papers are already available on ArXiv and published at conferences. Where can I download MAD-v2?

Soldelli commented 8 months ago

Dear @zzz-zq, MAD-v2 textual annotations are available for download. Check the "Request access to the MAD dataset" section of the README. You will need to apply to access the data and download the annotation file. Inside the tar file you will find both MAD-v2 and MAD-v2 annotations.

The visual features used by this work are the same used for MAD-v1. The language features are coming soon. However, you can obtain them easily by encoding them with the correct CLIP language encoder (ViT-B/32 or ViT-L/14, depending on which visual features you adopt).

Kindly, let me know if you have any other questions.

jianhua2022 commented 5 months ago

Hi, @Soldelli where to download the raw video data? I want to visualize some examples, but the downloaded data only contains video features.

Soldelli commented 5 months ago

Dear @jianhua2022, if you need the videos, you will need to purchase the movies through your institution. Sharing copyrighted material is illegal, and therefore, we cannot disseminate the raw videos.