When you select a model from the Model tab, you will see a grid view of thumbnails. This default view is called the Gallery view.
A model is an internal Labelbox representation of the machine learning model you select for your business use case. A model run represents a training experiment on that model. For each model run, Labelbox provides a versioned snapshot of all information (including the data rows) involved in the experiment at the time it was run. When you select a model, you can select from a dropdown list of model runs to display all of the data rows that are associated with each model run.
After you upload predictions to a model Run, you can use the gallery view to visually compare your model predictions against the ground truth labels. This qualitative inspection should help you gain a better general understanding of your model's behavior.
Here are the steps to easily visualize predictions in a model run:
Navigate to the Model tab
Select the model you want to visualize
Select the model run within the model that you want to visualize
- By default, you will see the model predictions for each data row in the gallery view. To come back to the gallery view, click on the gallery icon in the top right, above the thumbnails.
You can drill down into subsets of your data rows by narrowing down the data rows by data split. This allows you to visually inspect the annotations and predictions on a subset of the model run data.
For example, if you want to gain an understanding of your model's behavior on a particular data split, you can do so by filtering by Train, Validateor Test set. The gallery view will only show the data rows in the data split you selected. Clicking on All shows all of the data rows involved in the model run.
If you want to inspect your model's behavior on a specific class of annotations or predictions, you can filter by annotation or prediction name.
In the example below, the gallery view shows all data rows that contain an “airplane” annotation. You can further refine the filter by filtering all data rows that contain at least N “airplane” annotations.
Labelbox supports the following filters in the Model tab. You can combine these filters using AND and OR conditions.
- Data row ID
- Label ID
After you upload metrics to a model run, you can sort the Data Rows on these metrics. To do so:
- Click on Sort
- Select a metric to sort by
- Select an ordering: Ascending or Descending
All metrics in the model run are available for sorting:
- scalar metrics or confusion matrix metrics
- confusion matrix metrics are available as global metrics (value is averaged across classes) or as class-level metrics
You can combine filtering and sorting, in the grid view, to unlock powerful workflows. Learn more about the following workflows, to improve model performance, increase labeling quality and efficiency, and reduce labeling budget.
- Finding and fixing model errors.
- Finding and fixing labeling mistakes.
- Identifying high-value data to label in priority (active learning).
You can customize the display of the gallery view, by opening the Display panel. From there, you can customize the following display settings:
|Thumbnail size||Size of the thumbnails in the gallery view|
|Assets > Greyscale coloring||Color the data row in grayscale. This is useful to see annotations and predictions stand out|
|Features > Predictions > Show objects||Annotation objects (bounding boxes, entities, polygons, segmentation masks...) are visible/invisible on thumbnails in the gallery view|
|Features > Predictions > Show objects||Prediction objects (bounding boxes, entities, polygons, segmentation masks...) are visible/invisible on thumbnails in the gallery view|
|Features > Annotations > Show classifications||Annotation classifications are visible/invisible on thumbnails in the gallery view|
|Features > Predictions > Show classifications||Annotation predictions are visible/invisible on thumbnails in the gallery view|
|Object colors > Color by feature||All annotations and predictions associated with the same feature, have the same color. Annotations are represented by a solid line. Predictions are represented by a dashed line.|
|Object colors > Color by source||All annotations are represented in a solid green line. All predictions are represented by a dashed red line.|
To access the detailed view for a data row, click on any thumbnail. The detailed view allows you to inspect the details of a particular data row.
You can zoom in on the data with the mouse wheel.
The display panel allows you to customize the display settings below. You can access it by clicking on Display.
|Black color fill data row||Replace the data row with black pixels. This is useful to see annotations and predictions stand out|
|Greyscale||Color the data row in grayscale. This is useful to see annotations and predictions stand out|
|Color by feature||All annotations and predictions associated with the same feature, have the same color. Annotations are represented by a solid line. Predictions are represented by a dashed line.|
|Color by source||All annotations are represented in a solid green line. All predictions are represented by a dashed red line.|
In the detailed view of a data row, you can see attachments by opening the attachments panel.
See Attachments for instructions about how to create attachments and about supported attachment types.
Updated 3 months ago