microsoft / RegionCLIP

[CVPR 2022] Official code for "RegionCLIP: Region-based Language-Image Pretraining"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Pretraining dataset

Jimuyangz opened this issue · comments

Hi guys, great work!
I try to run the pretraining code but it seems the pretraining data is missing. I wonder if you plan to release the pretraining data, i.e., 'PATH_TRAIN' in 'RegionCLIP_RN50.yaml'. Or how can I preprocess the COCO dataset to enable pretraining? Could you provide a pre-processing script? Thank you so much!

@Jimuyangz Thanks for your interest in our work. As mentioned in #10, the pretraining code and configs have already been released. The pretraining data might not be provided in the near future due to company policy. Those datasets are all public and you can download and get it accepted by dataloader (e.g., the inputs are simply an image and a string representing a caption).