calvin & simpler finetune dataset #84
Description
Hello XVLA authors,
Thank you for opening up the XVLA codebase—it's an excellent work and has made an important contribution to the embodied intelligence community. I’m currently trying to reproduce your fine-tuning results on downstream simulation tasks, but I’m running into issues with dataset processing. I’d like to confirm your data processing workflow:
For the CALVIN dataset, which original data did you use? Is there a processing pipeline you can share, or any details/notes we should pay attention to?
In the paper, “Google Robot” is mentioned under Simpler. What specific data source is this referring to, and could you share the corresponding data processing pipeline? Also, for BirdGEv2, which exact dataset do you use?
If there are Hugging Face dataset sources and you’ve already processed them successfully, please let me know. Thank you very much.