SNAV fine-tuning setup

[Felipe-Tommaselli]

Congratulations on your work and acceptance at ICRA26!
Do you have plans to open-source the training and fine-tuning setup for your SNAV model? I appreciate your efforts in open-sourcing the evaluation code and data curation. Additionally, I would like to replicate your training recipe for SNAV with further ablations on the VLM side.

[TidalHarley]

Thank you so much for your interest in our work.
We do plan to open-source the training setup for SNAV once the code is properly organized and ready for release, and our current goal is to have it available within the next two weeks. After that, we also expect to share a more detailed version of our data augmentation pipeline beyond what is described in the paper. We will organize and release the full implementation details of the augmentation pipeline as soon as possible, although some parts may still require additional cleanup and consolidation on our side.
We would also be very happy to see more people participate in evaluating and analyzing the benchmark, and we warmly welcome follow-up work that builds on top of it.