search

xinyu1205 / recognize-anything

Open-source and strong foundation image recognition models.
https://recognize-anything.github.io/
Apache License 2.0
2.7k stars 255 forks source link

About training code #64

Open YahooKID opened 1 year ago

YahooKID commented 1 year ago

Halo guys， Thanks for your great work. Will your training code be released on time(7.8) and will it include the training code of RAM?

xinyu1205 commented 1 year ago

Hi, thanks for your attention! We have open-source Tag2Text forward function, which can refer to README.md. While I would personally love to open source all the training codes right away, they are still in the company's process. We are actively working towards making them available as soon as possible. Best wishes!

dydxdt commented 1 year ago

Hi, thanks for your attention! We have open-source Tag2Text forward function, which can refer to README.md. While I would personally love to open source all the training codes right away, they are still in the company's process. We are actively working towards making them available as soon as possible. Best wishes!

Hi, thanks for your great job! I want to ask if I replace 'blip.py' with 'ram.py', it also seems to be able to run(but I haven't tried), is that right? I see the code structure is similar.Thanks for reply

xinyu1205 commented 1 year ago

Besides, another key improvement is: BLIP only read two key-value pairs: {'image': path_of_image, 'caption': text_of_image}, and Tag2Text need to read three key-value pairs: {'image': path_of_image, 'caption': text_of_image, 'tag': the tags of the image, parsed from the caption}.

dydxdt commented 1 year ago

Thanks. So when will the ram training code be released? maybe an approximate time？

lichun-wang commented 1 year ago

非常赞的工作，非常期待训练代码～～