Open mingzeG opened 1 month ago
Hello, @yinanhe, how to download the CLEVRER dataset? The link shows noting.
@pengzhiliang
You can download from the link below.
Training Videos, Annotations, Questions and Answers
Validation Videos, Annotations, Questions and Answers
Testing Videos, Questions
Object Masks and Attributes
Readme
@yinanhe Thanks very much 🍻
@yinanhe Do you have alternative links? The videos links aren't working for me.
@schopra8 We no longer have any other links. If you are still having difficulties in getting the data, please let me know which dataset it is.
Apologies for the lack of specificity! I mean the CLEVRER dataset specifically. The question+answers and README load for me -- but the videos don't, when I click on those links.
I reached out to the original authors of the CLEVRER dataset as well, but they haven't responded.
@yinanhe -- I'm also struggling to find the VideoChat videos used in the conversation and caption annotations. Do you have any pointers to where I can download this data? Thank you in advance!
Apologies for the lack of specificity! I mean the CLEVRER dataset specifically. The question+answers and README load for me -- but the videos don't, when I click on those links.
I reached out to the original authors of the CLEVRER dataset as well, but they haven't responded.
@schopra8 how about use link in https://github.com/OpenGVLab/Ask-Anything/issues/176#issuecomment-2121805009. In my network environment, the download is normal.
@yinanhe -- I'm also struggling to find the VideoChat videos used in the conversation and caption annotations. Do you have any pointers to where I can download this data? Thank you in advance!
The videos used here are from YouTube, and some videos might be found in the InternVid dataset on opendatlab.com.
Thank you @yinanhe!
For CLEVRER it looks like my browser was automatically trying to turn the HTTP link into HTTPS and that was why the file was not downloading. This works for me now.
For the VideoChat portion of the VidChat2 Instruction Tuning data I see file names like "000551_000600/1054295129.mp4" but when I look at the InternVid-10M dataset in HuggingFace I only see YouTube Ids (e.g. "HdYoyzCSWyw"). How does one align the YouTube IDs to the names in the VidChat2 Instruction Tuning dataset?
I now see that the VideoChat data names correspond to WebVid like VideoChat2 -- and can resolve the videos.
@yinanhe - Thanks for all the help! It might be helpful to update the Data.md
file to clarify that VideoChat corresponds to videos from WebVid. The link to InternVid
threw me off -- and had me incorrectly looking at the InternVid-10M Dataset .
Great Work! I was hoping to quickly get the raw video dataset used and try to train a videochat2, how could I get a filtered raw video dataset rather than downloading all the video datasets that the data.md mentioned. It's too memory consuming