Update concepts with python

[phoenixsampras]

Hello,

I have a set of symbols in my application.

I've been looking for an example on how to UPDATE the image CONCEPTS and METADATA.

Please could somebody share some snippets.

Thanks in advance!

[jared]

Hi @phoenixsampras! To update inputs you can do the following:

One input at a time

update(image, action='merge')

Bulk update

bulk_update(images, action='merge')

For each of these:

images – an image or a list of Image objects that have concepts, metadata, etc.
action – one of [‘merge’, ‘overwrite’]
’merge’ is to append the info onto the exising info, for either concept or metadata
’overwrite’ is to overwrite the existing metadata and concepts with the existing ones


A full list of functions can be found at https://clarifai-python.readthedocs.io/en/latest/clarifai.rest/

Hope this helps!

[phoenixsampras]

Thank you for the response,

We have used this technique to create and assign concepts to the images, works fine.

        imagen_resultado = app.inputs.bulk_merge_concepts([imagen.input_id], [[(concepto,True)]])

Problem is that, the concepts that are created are not associated with any model, therefore when we try to use the concept we get this error. Also we cant find a way to Associate a Concept with a model.

http://prntscr.com/k8xi8n

Thanks in advance for your help

[jared]

Hey @phoenixsampras - to associate concepts with a model you can use this endpoint:

https://clarifai.com/developer/guide/models#add-concepts-to-a-model

Just out of curiosity though - does the error go away if you train the model after doing the bulk update?

[phoenixsampras]

Thanks for your response,

This is a research project, as there is no models for this kind of images. So we are testing with concepts, however the results are not as expected.

The error went away.

The results are not as expected, so I guess we give up on clarifai for now.

[jared]

Hi @phoenixsampras - sorry that your custom model isn't giving you the accuracy you are looking for! Curious to hear more about this in a private message if you'd like to share the details.