Congratulations on the release of your great work. I am wondering whether you have tried training the model on the unfiltered 800k Objaverse in your early exploration of the project? If so, dropping 800k to 50k and still achieving a better performance clearly highlights the importance of training data quality. Considering that, would that be possible for you to kindly release the 50k object ids / filtering code which I believe would be a great contribution to the 3Dgen community.
Dear Authors,
Congratulations on the release of your great work. I am wondering whether you have tried training the model on the unfiltered 800k Objaverse in your early exploration of the project? If so, dropping 800k to 50k and still achieving a better performance clearly highlights the importance of training data quality. Considering that, would that be possible for you to kindly release the 50k object ids / filtering code which I believe would be a great contribution to the 3Dgen community.