Commit
•
990522e
1
Parent(s):
9adba38
Update README.md
Browse files
README.md
CHANGED
|
@@ -4,4 +4,13 @@ task_categories:
|
|
| 4 |
- text-to-image
|
| 5 |
size_categories:
|
| 6 |
- n<1K
|
| 7 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
- text-to-image
|
| 5 |
size_categories:
|
| 6 |
- n<1K
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
# NASA Earth Instagram
|
| 11 |
+
|
| 12 |
+
This dataset is a moderately curated subset of the posts shown on [NASA Earth's Instagram](https://www.instagram.com/nasaearth/), with an emphasis
|
| 13 |
+
on finding image-text pairs where the text associated is as close as possible to being a direct caption of the image in question.
|
| 14 |
+
|
| 15 |
+
This dataset has a variety of use cases, but the one which it is originally intended for is to provide a fine-tuning dataset for image captioning models,
|
| 16 |
+
to be better equipped for describing the exact pheonomena in satellite imagery.
|