Datasets:
Issues with missing splits and invalid file
Hi, thanks for sharing this dataset!
I encountered a couple of issues while trying to use it:
Missing files
It seems that some splits are incomplete. In particular, the medium and hard subsets for the No obstacle configuration appear to be missing.
Invalid .npy file / loading errors
When working with the file: FPO_Geometry_Medium_SingleObstacle.npy
- Using np.load(..., allow_pickle=False) raises:
ValueError: Cannot load file containing pickled data - Even with allow_pickle=True, I get:
Failed to interpret file as a pickle
This suggests that the file may be corrupted or not a valid .npy file.
Any clarification or fixes would be greatly appreciated. Thanks again!
Hi, thanks for the kind words and for taking the time to report these issues! Apologies for the late reply, it's been a while since we left CMU and it took us some time to track things down.
Corrupted file (FPO_Geometry_Medium_SingleObstacle.npy): We've confirmed the file was corrupted and have uploaded a fixed version, please use Geometry_Axis/FPO_Geometry_Medium_SingleObstacle_v2.npy instead. It's a valid .npy file (shape (6284, 20, 128, 128, 6), dtype float32) and should load without issues using np.load(..., allow_pickle=False).
Missing files (No obstacle, medium & hard splits): We'll look into locating those files, apologies for the inconvenience. In the meantime, we've included data generation scripts in the repository precisely for cases like this. The No obstacle configuration is one of the simpler setups and shouldn't take too long to regenerate if needed urgently. We'll update here once we have more information.
Thanks again for your patience and for helping us improve the dataset!