Does “multilingual MSCOCO” refer to “XTD”?
#38
by
Chao985
- opened
In the technical report of Jina-CLIP-v2, the model's performance on the "multilingual MSCOCO" dataset is mentioned, while the cited paper is the original MSCOCO dataset paper. I wonder if the "multilingual MSCOCO" here refers to the XTD dataset.
Hello @Chao985 ! That's a good catch! I remember this issue was raised at some point because we couldn't find a reference for Multilingual MSCOCO but then we probably forgot to resolve it. You are right, the dataset is XTD10, the details are here https://github.com/LAION-AI/CLIP_benchmark/blob/main/clip_benchmark/datasets/multilingual_mscoco.py