microsoft / GLIP

Grounded Language-Image Pre-training
MIT License
2.06k stars 186 forks source link

flickr annotation? #72

Open george2099 opened 1 year ago

george2099 commented 1 year ago

Thanks for your greate work. I have a question about the final_flickr_separateGT_train.json you provide, that one caption have several annotation for same category, like this:

4657935d127133fdd92a2e3c852ebaa6

for training, will the dot_product of visual and bert embedding cause false negative pair? @Haotian-Zhang @liunian-harold-li

jiaxi0030 commented 7 months ago

Sorry to bother you. Is it necessary to divide the data set into val\test when evaluating the flickr30k data set?