Skip to main content

⚠️ Please note that this topic or post has been archived. The information contained here may no longer be accurate or up-to-date. ⚠️

I made a Machine Learning based automatic image tagger for CaptureOne.

Sam Mullins

I made a Machine Learning based automatic image tagger for CaptureOne. It's currently a proof of concept but is this something people would be interested in me developing further? It works completely offline so there are no charges or usage limits.

https://www.youtube.com/watch?v=mb-dPQU-3Tc

If it was something people were interested in, we could community source a whole range of images/categories to train the model with. Currently it tries to detect the specific type of thing whereas in real life usage it would be useful to have model that returns "animal, big cat, tiger" or "landscape, golden hour, barn" rather than just "tiger" and "barn".

1

Comments

6 comments

Date Votes
  • André Maurer

    Looks very good. Capture One is very slow with new technologies. Would like to see some AI things or even HDR. 

    Work your Plugin also on a Mac?

    0
  • Sam Mullins

    Unfortunately I don't have a Mac, so I won't be able to build a version for that. I'm in the process of building custom image classification machine learning models so that the labels are more appropriate for tagging photos into categories. My code will be open source too, so someone willing to make the Mac version would have a head start once I was finished.

    Re. HDR... It may not be useful for you, as again it's Windows only, but one of my other open source CaptureOne plugins (PanoCapture) can make very basic HDR and panorama images with a single button click. 

    0
  • André Maurer

    So I hope that somebody does that for a mac. 

    I wanted to make a bachelor students project, but the SDK is really week and different from windows to mac. But sadly Capture One doesn't care. 

    Happy to read more about your future projects and followed you on youtube.

     

    1
  • Anthony DE FARIA

    Wow! I like this kind of hack, well done!

    I'm thinking of doing something like this but with a different approach maybe. I thought about using an image tagging API to generate tags. It will not be free but I won't have to train a model... Then I'll have to send back the information to capture one. I have no idea how to do that but I can see that you did it.

    Have you thought about transfer leraning? maybe you can save some time with your model training.

    I'm on a mac by the way ;)

    0
  • Sam Mullins

    Yeah I have done a transfer learning version and seemed to work pretty accurately. Even with the transfer learning method you need to source and sort a whole load of images and I was struggling with finding a good multi label model to use. I had a single label one working fine, but I didn't find only tagging with one label to be that useful.

    0
  • Anthony DE FARIA

    Maybe you can scrap shutterstock like websites. A thumbnail should be enough for the model and you will have a load of labels. One example here:

    Here one example with a lot of keywords:

    https://www.shutterstock.com/fr/image-photo/background-elephant-133689230

    0

Please sign in to leave a comment.

Didn't find what you were looking for?

New post