Evaluation on ReasonSeg test set (missing annotation, text for refer)

[suikei-wang]

Thanks for the amazing work!

I am currently running some evaluation on ReasonSeg test set and I found that there are 2 jsons files are missing the polygon shape. I simply removed them from the test set (now 777 images). Not sure if this affect the evaluation result.

Also I am wondering how you take the texts for an image when you perform the evaluation? Do I need to take each text in list texts as a refer? Apparently all texts are referring to the same region (they share an annotation) but I am not sure if the evaluation are done for each text.

Thank you!