client
client copied to clipboard
log_image support?
Similar to https://www.tensorflow.org/api_docs/python/tf/summary/image or https://mlflow.org/docs/latest/python_api/mlflow.tracking.html#mlflow.tracking.MlflowClient.log_image
Thanks @elgalu , that is a suggestion we've been thinking about how to implement properly.
Why not just save the image as part of the commit, next to the output metrics.csv
and params.yml
?
You expect to see the image as part of the experiment dashboard in the end?
You expect to see the image as part of the experiment dashboard in the end?
Yes, even more, we expect to see each logged image in each step to track model evolution
Interesting, I would be happy to hear more if you can share.
- Do you have an example of a different tool that does as a reference?
- Can you describe the use case? Why is this valuable to you and how you expect to use this? Details would be valuable, e.g. if you only actually look at 2-3 images and not for every step, or only once every epoch vs. once every step
- Can you describe how you imagine using this feature in the UI?
In the meantime, why not save images to a DVC tracked folder with the step index as part of the image name, as a workaround?