bop_challenge_2023_training_datasets.md

Training datasets for Tasks 4–6 of BOP Challenge 2023

The datasets include over 2M images showing more than 50K diverse objects. The images were originally synthesized for MegaPose using BlenderProc. The objects are from the Google Scanned Objects and ShapeNetCore datasets and their 3D models can be downloaded from the respective websites.

Note that symmetry transformations are not available for these objects, but could be identified using these HALCON scripts (we used the scripts to identify symmetries of objects in the BOP datasets as described in Section 2.3 of the BOP Challenge 2020 paper; if you use the scripts to identify symmetries of the Google Scanned Objects and ShapeNetCore objects, sharing the symmetries would be appreciated).

MegaPose-GSO dataset

• 3D object models can be downloaded from Google Scanned Objects. For the models to be compatible with GT poses, you will need to (i) center the objects and rescale them such that they fit within the unit sphere, and (ii) scale the normalized models by 0.1. Please see this comment for pseudo code.
• Mapping from obj_id used in BOP to the original object identifiers
• Mapping from an image key to the index of the shard where it is stored
• The dataset is in the BOP-webdataset format and is split into 1040 shards, with each shard containing ~1000 images together with object annotations and camera parameters. Use the following URL template to download a shard (<SHARD-ID> is from 000000 to 001039).

https://huggingface.co/datasets/bop-benchmark/megapose/tree/main/MegaPose-GSO/shard-<SHARD-ID>.tar

MegaPose-ShapeNetCore dataset

• 3D object models can be downloaded from ShapeNet (scale the models by 0.1 to be compatible with GT poses)
• Mapping from obj_id used in BOP to the original object identifiers
• Mapping from an image key to the index of the shard where it is stored
• The dataset is in the BOP-webdataset format and is split into 1040 shards, with each shard containing ~1000 images together with object annotations and camera parameters. Use the following URL template to download a shard (<SHARD-ID> is from 000000 to 001039).

https://huggingface.co/datasets/bop-benchmark/megapose/tree/main/MegaPose-ShapeNetCore/shard-<SHARD-ID>.tar