d-cube
d-cube copied to clipboard
# of negative samples in the intra scenario
Thank you for the great work!
I want to clarify the following statistic in the paper:
For the intra-scenario setting (where candidate descriptions for an image only come from the same scenario), there are 20,279 positive pairs and 53,383 negative pairs.
I listed all the image-text pairs with no target (within intra-scenario) and found a total of 29,067. Is the 53,383 referring to the number of object-sentence pairs rather than image-sentence pairs?