No features found when processing Monado SLAM Dataset

[RiccardoParosi]

Hi, I tried running sai-cli process on the Monado SLAM Dataset at https://huggingface.co/datasets/collabora/monado-slam-datasets, but no features are detected and therefore no estimated trajectory is generated.

This dataset is a collection of egocentric visual‑inertial sequences from VR headsets, structured in a EuRoC‑like layout. I converted the sequences to SpectacularAI format using sai-cli convert tum, and I used intrinsic and extrinsic parameters provided by the dataset. I tried both with and without the --fast flag.

What should I modify so that the dataset is processed correctly?

[oseiskar]

Hi @RiccardoParosi . Sorry for the delayed reply. We unfortunately cannot offer (non-commercial) support for this kind of troubleshooting at this point. There are several things that can be wrong. The usual suspects include:

• coordinate conventions: imu-to-cam vs cam-to-imu, local camera coordinate systems. You can check our conventions here https://spectacularai.github.io/docs/sdk/tracking.html#coordinate-systems
• calibration coefficient order and models (see here, page 9: https://spectacularai.github.io/docs/pdf/calibration_manual.pdf )
• timestamp units and other sync issues