LAVIS
LAVIS copied to clipboard
Modify BLIP2 pretraining config files to add more dataset
Thanks for the great work!
The BLIP-2 paper mentions that the model is pre-trained on a combination of dataset, including COCO, Visual Genome, CC, SBU and LAION.
Looking at the provided config files, however, it seems only coco_caption
and vg_caption
are used.
To produce the results, do we need to modify the config files and add the other dataset?