Tagging

In Tenyks, tags 🏷️ are labels assigned to categorize and identify related items, such as images and objects.

📘

Objects

We define an object as a labelled instance of a specific class. An image can contain one or more objects. Each object is defined by a bounding box (object detection) or by a mask (segmentation).

The goal of tags is that you can easily organize, search, retrieve and evaluate items that have been tagged with the same label or keyword.

To illustrate this feature, we use a medical domain dataset. If you're interested in analyzing this dataset, 👉 try it here on Sandbox!

1. How to tag images

To tag images follow the next steps (see Figure 1):

1. Click on "Select", located on the top left side of the Data Explorer
2. Two options will show up: Image or Object. Choose "Image"
3. Click on the images you want to tag (you will see a square 🟦 around the borders of a selected image)
4. On the top right side of the Data Explorer, click on "...More"
5. Select "🏷️ Tag"
6. Provide a name and a value for your tag, following this format tag_name:tag_value. Be aware that a semicolon ":" is required between "tag_name" and "tag_value".
   1. A tag named annotation_errors can have values such as mislabelled, etc.
7. Click on Create

2. How to tag objects

To tag objects (Figure 2), follow the same steps highlighted in "How to tag images" except that in step 2, instead of Image, select Object.

3. How to remove tag names or tag values

To delete tags (Figure 3), follow the next steps:

1. On the horizontal menu, click on Image Tags or Image Objects
2. A list containing all the "tag names" will be displayed
3. Hover over one of the "tag names" to reveal the values associated with this tag
4. Remove a "tag value" or an entire "tag name" by clicking on the cross mark button

4. What you can do with tags

Once you have tagged some images or objects, here are some ways you can use them in your workflows:

4.1 Filter by tag

The most basic functionality for tagged images or objects is to use them to filter your dataset as shown in Figure 4.

• Scenario: A model that predicts tumours of different kinds might be failing to detect a particular class.
• Tags: Not all the failing predictions are likely to reflect the same story, hence you might be interested in creating tags that break down your undetected objects as follow:
  • age: pediatric, adult, geriatric
  • stage_of_disease: stage_I, stage_II, metastatic
  • previous_treatments: chemotherapy, surgery

Using tags you can provide structure to your data filtering ⭐.

4.2 Evaluate model performance on a set of tagged images

Tagged images or objects can be used to group data points that we are interested in evaluating model performance on, see Figure 5.

• Scenario: You suspect that your model is failing to detect some particular edge cases. How can you evaluate performance on these outliers in minutes rather than in days?
• Tags: Let's assume the edge cases are related to the size of the medical condition.
  • First, you can create some tags describing this scenario. For instance, tag name: size, values: small, medium.
  • Second, you can evaluate model performance on these two groups of edge cases by simply selecting them on the left side of the Model Comparison feature.

The time to perform this operation takes minutes! Tenyks really 10x your work as ML engineer! 🚀

4.3 Use tags to filter by colour on the Embedding Viewer

Filter the embedding space, in the Embedding Viewer, based of your dataset by using tags as reflected in Figure 6.

These are only some of the different ways you can use tags on the Tenyks platform.