Hi OpenSearch-VL team,
Thanks for the great work!
I noticed that the README mentions the planned release of the data curation pipeline, including Wikipedia path sampling, fuzzy entity rewriting, and source-anchor visual grounding.
I’m very interested in this part, since it seems crucial for understanding how the SearchVL SFT/RL data was built and for extending the method to new multimodal search tasks.
A preliminary version, partial scripts, or even the expected input/output data schema would already be very helpful.
Thanks again for the excellent work!
Hi OpenSearch-VL team,
Thanks for the great work!
I noticed that the README mentions the planned release of the data curation pipeline, including Wikipedia path sampling, fuzzy entity rewriting, and source-anchor visual grounding.
I’m very interested in this part, since it seems crucial for understanding how the SearchVL SFT/RL data was built and for extending the method to new multimodal search tasks.
A preliminary version, partial scripts, or even the expected input/output data schema would already be very helpful.
Thanks again for the excellent work!