I want to create a very basic synthetic dataset (similar to NDDS) that I can pass to PVNet. Since PVNet is very old, and a bunch of its infra is EOL (e.g. Ubuntu 16.04) I was not able to setup PVNet Rendering. Further, I saw one user using NDDS and with some modifications to basecode, she was able to detect pose. However, NDDS seemingly requires UE4.22 which is not anymore downloadable from Unreal Engine (I have used git private token and I am part of UE org but the AWS file for that download is not available anymore – please correct me if I am wrong). Further, I couldn’t find any guide for using UE5 for NDDS build and it seems NDDS (from NVIDIA) is out of git support and is currently unsupported (no further responses to git issues). Also, I am wondering, if you may have any internal leads from NVIDIA if NDDS is buildable with UE5 in Ubuntu 22.04?
Since PVNet is a highly cited and brilliant paper in the field and still a couple other frameworks base their method on it (e.g. HybridPose), I am curious if you could please provide a methodology or script for creating a basic synthetic dataset for a single object of interest (that is not in famous BOP challenge datasets such as linemod, tless) so that I could train PVNet with it?
I also saw this post but it seems there was no extension/script for creating a 6DoF pose estimation dataset for a single object of interest and it would be very valueable for the community if we could have datasets in linemod format since it is the most used format for BOP challenge which is the predominant challenge in 6DoF pose research.
Further, I would suggest as a longer mode plan of code release to add the feature of Linemod/BOP for Writer such as the current one DOPEWriter. Thanks a lot.