About GranD Pre-training Dataset about groundinglmm HOT 2 CLOSED

Hi @Shengcao-Cao,

Thanks a lot for reaching out and for your interest in our work!
Yes, our dataset will include all 11 million images from the SA-1B dataset. We plan to release 1-2 million images weekly. It's been challenging to get to this point dealing with limited resources, having to redo some of our annotations because of data loss, and improving our models. This has made things take a bit longer than expected. The run-pipeline.sh script shows how complex our models are, which explains the delay in releasing everything.

We have released all the code necessary for generating the dataset (run-pipeline.sh). If you have the resources available, we encourage you to explore and utilize this code. We will try to help with any reproduction efforts.

Sorry for the wait, and thanks again for your support!

from groundinglmm.

HI @hanoonaR ,

Thank you very much for your reply! I totally understand the complexity in your data preparation. Sincerely appreciate your efforts in sharing this great work with the community.

Best,
Shengcao

from groundinglmm.