feedback loop azure ml possibilities?

Question

feedback loop azure ml possibilities?

Hamza Outa 61

for a project we are doing azure ml assisted labeling (object detection). we also want to check the model for false positives/negatives, by receiving the results of the camera and performing manual labeling on it. then we want to take the results from the check and put those in the original dataset that we created using azure labeling. is there a built in function for this or do we need to create this ourselfs?

romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2023-02-16T07:54:46.0333333+00:00

Hello Hamza Outa,
Just checking in to see if the below answer helped. If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Accepted answer

0 additional answers

Your answer

romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2023-02-16T07:54:46.0333333+00:00

Hello Hamza Outa,
Just checking in to see if the below answer helped. If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Answer 1

romungi-MSFT 48,911 Microsoft Employee Moderator

Hamza Outa Are you looking to add new labels after some feedback for your systems?

I think there is an option to add new labels to a project by pausing it and then you have the following options:

Start over, removing all existing labels. Choose this option if you want to start labeling from the beginning with the new full set of labels.
Start over, keeping all existing labels. Choose this option to mark all data as unlabeled, but keep the existing labels as a default tag for images that were previously labeled.
Continue, keeping all existing labels. Choose this option to keep all data already labeled as is, and start using the new label for data not yet labeled.

Is this what you are looking for? I think the second option might work for you to add new labels if the default tags need to be changed. As per documentation you can start using the new labels for labeling and the ML assisted labeling will start after a certain threshold is reached or you can manually start an ML assisted training run. I hope this helps!!

Hamza Outa 61 Reputation points

2023-02-16T09:40:16.38+00:00
no I am trying to fine tune a model by updating the dataset (labels stay the same).

let's say i have a model that was trained using the dataset that we exported from data labeling.

that model is running on a device.

what i now want to do is: take some of the footage from that device and upload it to the same labeling project to validate the labels and correct if wrong.

is there a built in function for this or do i need to create this functionality myself?
romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2023-02-16T10:40:29.99+00:00

Ok, are there two scenarios here? Have you used the labeling project to label your dataset which was then exported as an Azure ML dataset to create a model on Azure ML workspace?

If Yes, then you can add additional files to your dataset if you have setup incremental refresh to add new files to the project. After adding new files, you can continue to run manual or ML assisted labeling and export the new dataset to train your new model in the workspace. This activity of exporting new dataset and retraining is not automated, you will have to follow the same steps as before and continue to create newer models.

If you have created a pipeline though of your model then this can be automated to an extent with Azure ML pipelines.
Hamza Outa 61 Reputation points

2023-02-17T14:52:26.2333333+00:00

oh ok thanks, that clarifies it.

can i also upload labeled images (what my model made) to the dataset and edit those?
romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2023-02-20T06:36:03.07+00:00

Currently, the labeling projects cannot import labeled images.

Share via

feedback loop azure ml possibilities?

0 additional answers

Your answer