about the precomp dataset

[lt152]

Great job! However, I have a question to ask for clarification. When I downloaded the dataset from Kaggle, I found that the dev images in the Flickr 30k dataset contain 5070 images. Do they actually represent 5070/5 = 1014 unique images (one image corresponding to 5 texts)? In the COCO dataset, the dev images amount to 5000, does this mean they represent 5000/5 = 1000 unique images? This seems to contradict the original statement of having 1000 validation images for f30k and 5000 for COCO.