zerovl

@rwightman Thanks for replying. We agree with that the logic should be placed in a dataset wrapper. We are working on implementing it, and making sure experiment results are correct.

@rwightman hi, the implementation is done, and the log is attached. It seems that results are almost the same with the former version. Would you check the code when you...

@rwightman thanks for your time. I am willing to discuss about implementation details.

@rwightman the implementation details have been wrapped into a separable class. However, I am not familiar with the 'TrainJig' class. Would you check if the code works well with 'TrainJig'...

> Closing for now since it's supported, thank you! @mitchellnw Thanks for your efforts! Would you think it is okay to mention our [eccv'22 paper](https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136960234.pdf) about the gradient accumulation with...

@mitchellnw greatly appreciate your updating!

Yes, could you leave your email address or something else, and I will send files to you. BTW, i need to clarify that you would better create the annotation files...

https://pan.baidu.com/s/1EIPztHFLjeV_03AbnlvDZQ?pwd=43gn you will find annotation files of coco, f30k and sbu in this url. I will update annotation files of other datasets in the following days.

Hi. The image_id and caption_id can be randomly given. I rename the image files, so the image_id is the same to the image file name. As I said before, you...

my pleasure！ From: ***@***.***> Date: 2月28日 (周一) 23:00 Subject: [External] Re: [zerovl/ZeroVL] train_anno.csv and val_anno.csv (Issue #2) To: ***@***.***> Cc: ***@***.***>, "Comment"< ***@***.***> Thanks! — Reply to this email directly,...