Missing dependencies for SFS conversion pipeline

[TharitSinsunthorn]

Component

Dataset / Zarr / RLDB

Description

When I try to run the dataset converting script from sfs format to Zarr by sfs_to_egoverse_zarr.py, I got a ModuleNotFoundError including with scale-sensor-fusion-io and decord.

Should we add these packages to requirement.txt?

Thank you!

Steps to Reproduce

python sfs_to_egoverse_zarr.py --task-ids <TASK_ID> --output-dir ~/EgoVerse/coc/flash7/scratch/scaletoZarrDataset

Error Log / Traceback

ModuleNotFoundError: No module named 'decord'

and

ModuleNotFoundError: No module named 'scale_sensor_fusion_io'

Configuration

Environment

• OS: ubuntu 22.04
• Python 3.11
• PyTorch 2.7.1+cu128

[SimarKareer]

Hi @TharitSinsunthorn! I'm first wondering why you're running sfs_to_egoverse_zarr. This script converts from the raw ScaleAI data format to zarr, so this script is mostly for ScaleAI to use internally.

All the processed zarrs for their data already exist, and can be pulled (follow zarr_data_viz.ipynb).

Does this solve your problem?

[TharitSinsunthorn]

@SimarKareer Hello, thank you for your reply.

Currently, I’m working on a project that may use data from various sources, and ScaleAI data is one of them. I’m trying to see whether it’s possible to convert SFS data to Zarr format and then train it using EgoVerse’s scripts

I have tried following zarr_data_viz already, I managed to run through the code. However, there are some minor issues such as the updated of attributes inside the current version of the main branch that cause error.

For example, if I change
key_map = Aria.get_keymap(mode="cartesian") -> key_map = Aria.get_keymap(keymap_mode="cartesian")

This change work for me. If it is better to open a new issue please let me know.

If I’ve misunderstood anything, I would really appreciate your guidance.

[SimarKareer]

Ahh very cool, that makes sense!

If you want to make a PR adding those missing packages and this fix to zarr_data_viz I'll merge it in asap.

[TharitSinsunthorn]

OK sure! I will create the PR, please check it whenever you have time.