I noticed that you updated the train_pbr data 6 days ago. To ensure my understanding is correct, for data downloaded before March 20th, is it sufficient to just replace the original annotation files with those from the updated ipd_train_pbr_patch.zip? Or do I need to redownload all the training data?
Your assumption is correct, downloading the patch and unzipping it into existing ipd folder so that it merges with train_pbr is enough!
Thank you for your response. After visualizing and comparing the datasets, I’ve confirmed that the corrections were specifically made to the poses of the hex_manifold object.
To optimize my workflow, could you please confirm whether this update exclusively addressed hex_manifold without modifying other objects’ data? This information would help me significantly save time by avoiding unnecessary retraining on unaffected objects.
I greatly appreciate your guidance and look forward to your clarification.
It should affect objects obj_000001 and obj_000011 (one of them is hex_manifold, another is one of the brackets), in scenes 30 to 49
Thank you for your reply!
@vaheta is still on the dataset. I see 30GB+ how do we work on them?
And I saw in the walkthrough readme that the total size is 6gb if downloading using the bpc toolkit. Could you clarify this?
Lastly, what compute resource can one use for this competition? Can it be run locally or on a PC with about 16gb GPU?
I see 30GB+ how do we work on them?
could you please clarify, what do you mean?
And I saw in the walkthrough readme that the total size is 6gb if downloading using the bpc toolkit. Could you clarify this?
by default the bpc CLI tool doesn’t download train and test data (which are >200gb combined). The remaining data is much smaller. You can run bpc fetch ipd_all to download everything.
Lastly, what compute resource can one use for this competition? Can it be run locally or on a PC with about 16gb GPU?
Once you submit the solution, the evaluation is done on our servers. For training and experiments you can use any resource you have