Visualizing predictions
Updated at January 8th, 2025
📘Note
Sama Vision is currently in Beta. It is a third-party tool we are using to visualize data, and in this case to visualize pre-annotations
About predictions
When a model is available on your project, you will be able to see shapes on your data directly in Voxel before the tasks are sent for annotation. As you can see in the visual below, it will show you your assets and predictions together.
The goal of pre-visualizing predictions
This tool is a key step because it enables you to quickly flag if the predictions were wrong, or if they are useful. It is meant to be a safety checkpoint before we send our tasks to associates with pre-annotations.
How to verify predictions
First, select a batch ID on the left side. Once you have it selected, you will see a random sample of tasks from this batch. Look through at a high level, to see any major prediction discrepancies.
Here are some example scenarios to look for:
One gigantic bounding box is taking the entire space |
No bounding box at all
|
|
A large quantity of bounding box overlapping |
A large quantity of bounding box with weird shapes
|
|
What to do after verification
Once you’ve verified the predictions, you are ready to send this data to the hub. To do so, click on the button ‘’submit annotation’’ and it will send the selection for annotation on the hub.
Once you click, you will have the option to decide if you want to send this batch with pre-annotations, or without pre-annotations