Hello,
I am currently trying to train a Keypoint Detection model using Roboflow . I used a dataset that I downloaded from Roboflow. When I look at the version tab, I see that there are 635 images (which makes sense with the dataset I used:
However, when I look at the dataset tab, I see 267 images.
Which one is accurate? What does each number reflect?
hello 
if you are looking at your dataset tab in universe.roboflow.com it might be because the rest of the images are split in either valid or test.
another possibility might be augmentations have been added. that the original number of the dataset is 267, but when creating a version you can choose to add augmentation to further increase your training data.
if you want me to look over it feel free to add the link 