search

Soldelli / MAD

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

what are the movie names of MAD training set? #8

Closed fake-warrior8 closed 1 year ago

fake-warrior8 commented 1 year ago

Hi, I did not find the movie names for the MAD training set in the annotation files, the paper or the supplement file, there are only movie IDs in the annotation files. Could you release the original movie names of the MAD training set? I found some studies like Guidance processed the raw videos.

Soldelli commented 1 year ago

Dear @fake-warrior8 thank you for reaching out. You can use this file which maps the dataset's IDs to the imdb identifier. If you need any movie information you can directly scrape it the related imdb page composing the url as https://www.imdb.com/title/IMDB_ID/.

Feel free to reach out if you have any doubt.

fake-warrior8 commented 1 year ago

Dear @fake-warrior8 thank you for reaching out. You can use this file which maps the dataset's IDs to the imdb identifier. If you need any movie information you can directly scrape it the related imdb page composing the url as https://www.imdb.com/title/IMDB_ID/.

Feel free to reach out if you have any doubt.

Thank you for your reply.

fake-warrior8 commented 1 year ago

Dear @fake-warrior8 thank you for reaching out. You can use this file which maps the dataset's IDs to the imdb identifier. If you need any movie information you can directly scrape it the related imdb page composing the url as https://www.imdb.com/title/IMDB_ID/.

Feel free to reach out if you have any doubt.

Can you share with me the raw movie videos or can you provide a way for me to download these movie videos, so that I can train some models using the raw videos. I promise these videos are for academic research use only.

Soldelli commented 1 year ago

Dear @fake-warrior8, unfortunately, that is not possible. I hope you will understand that these videos are protected by copyright, which is legally binding. If you were to need the videos, you would have to purchase the movies through your institution.

Thank you for understanding.

fake-warrior8 commented 1 year ago

Dear @fake-warrior8, unfortunately, that is not possible. I hope you will understand that these videos are protected by copyright, which is legally binding. If you were to need the videos, you would have to purchase the movies through your institution.

Thank you for understanding.

Thank you for your reply. could you please share the downloading script, I will download these movies after purchase these videos.

Soldelli commented 1 year ago

Dear @fake-warrior8 I am not sure I understand your request.

fake-warrior8 commented 1 year ago

Dear @fake-warrior8 I am not sure I understand your request.

I mean, how to purchase and download these videos in batch, instead of purchasing and downloading these movies one by one from the imdb website.

Soldelli commented 1 year ago

Dear @fake-warrior8, there seems to be some confusion. We did not purchase the movies on IMDB. We only crawled metadata information from there. We used this for drawing some statistics reported in the paper. There is no script for batch-purchase of such data. I hope this clarifies. Feel free to reach out for any other inquiry.

fake-warrior8 commented 1 year ago

Dear @fake-warrior8, there seems to be some confusion. We did not purchase the movies on IMDB. We only crawled metadata information from there. We used this for drawing some statistics reported in the paper. There is no script for batch-purchase of such data. I hope this clarifies. Feel free to reach out for any other inquiry.

So where did you download the raw video? Did you download these movies in batch?

Soldelli commented 1 year ago

Dear @fake-warrior8, movies were not "downloaded". Additionally, the process was not done in one go but split into several phases for the proof of concept before reaching the current dataset size. No one source allows obtaining the videos in "batch". However, if you do find any that is particularly convenient, I invite you to share it with me. I will relay the message to other interested parties.

bofang98 commented 8 months ago

We only crawled metadata information from there. We used this for drawing some statistics reported in the paper.

Dear @Soldelli, Thanks for your interesting work. By the way, can you share how you crawled the metadata information if you do not have the raw movies? Specifically, how do you extract the video CLIP features without having movies or their frames?

Soldelli commented 7 months ago

Dear @bofang98, we are in possession of the raw movies from which we extracted frames and used the official CLIP implementation (link) to compute the features we released. We additionally crawled metadata from IMDB simply for statistic purposes. We do not release nor maintain such crawling script.