Datasets:
Commit
•
7eefbbf
1
Parent(s):
6d75d74
create README
Browse files
README.md
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
This dataset is created from subset of [Conceptual Captions](https://ai.google.com/research/ConceptualCaptions/). The original dataset has 12M captions but this dataset has around 10M image, caption pairs in different languages with 2.5M unique images. This dataset has captions translated from English to Spanish, German, French using language specific English to [Marian](https://huggingface.co/Helsinki-NLP) models (with sequence length 128). Data distribution is following:
|
| 2 |
+
|
| 3 |
+
`train_file_marian_final.tsv`: 10002432 captions (2500608 captions of English, German, Spanish, French each)
|
| 4 |
+
<br />
|
| 5 |
+
`val_file_marian_final.tsv`: 102400 captions (25600 captions of English, German, Spanish, French each)
|