How to curate fine-tune dataset for SAM2?

As said in document, SA-V is a special dataset format dedicated for SAM2

https://github.com/facebookresearch/sam2/tree/main/sav_dataset/example

However, It seems that the Roboflow, Label Studio and so on tools can't support SA-V dataset format?

It's a little complicated to convert Bounding Box format of Label Studio JSON-MIN to SA-V RLE format

Anybody can tell me how do you curate dataset to fine-tune SAM2?

At least , tell me if there is library can convert BB to RLE?

Start with the SA-V dataset format and example linked in the issue, then compare them with the Label Studio JSON-MIN bounding-box format. Done would be clear documentation of the dataset-curation workflow, including whether a supported conversion library or method exists for producing SA-V RLE data.

Written by the indexing model from the issue text.