So many wrong captions in the dataset.
#4
by
wusize
- opened
There are so many 'a train station in city' in input_captions and output_captions, which are obviously inconsistent with the images. For example, the sample with idx=5 and split='test'.
And can I know which specific CLIP and DINO models were used to evaluate the performance?
I met the same question, I don't know if I should skip these samples.