salesforce / LAVIS

LAVIS - A One-stop Library for Language-Vision Intelligence
BSD 3-Clause "New" or "Revised" License
9.19k stars 909 forks source link

blip2 pretain dataset #469

Open jingwang97 opened 11 months ago

jingwang97 commented 11 months ago

hi,it seems like that the dataset of pretrain stage1 and stage2 mentioned in the blip2 paper contains coco,cc3m,cc12m,sbu and laion ,but the config file only include coco and vg dataset.which is true data using in the pretraining stage?

dxli94 commented 11 months ago

We use datasets mentioned in the paper.

The config file is an example. You need to add datasets you need.