Question About Missing Images in the Training Split
#2
by Joe-Hall-Lee - opened
Hi, thanks for sharing this great dataset!
I noticed that the training split seems to have incomplete images. The JSON file references 35,428 images across 3,565 samples, but only 9,985 images are available in train_data/images/. For example, sample 16783992093428003941 has 10 image paths (_0 through _9), but only _1 and _8 exist — the indices are not contiguous.
Are the remaining training images missing from the upload, or is this intentional? If they exist elsewhere, could you point me to them?
Thanks!