Autotag Queries
Autotag Query - Examples
“Show me all images tagged by my “rain” Autotag”
autotag.image = rain
autotag.image.rain > 0
“Show me all images where the “rain” Autotag score is above -0.5” (all Autotag scores range from [-1, 1], and 0 is the default threshold)
autotag.image.rain > -0.5
“Show me all images with stop signs tagged by my “stop_sign” Object Autotag”
autotag.object = stop_sign
Autotag Query - Details
When you create an image Autotag in Nucleus, a score for that tag gets applied to every image in the dataset, ranging from -1 to 1, with higher scores indicating greater confidence. You can see the distribution of scores for any Autotag by looking at the Insights tab. When you create an Object Autotag, a score is similarly applied to each bounding box in the dataset (including all ground truth bounding boxes, and all bounding boxes from each model run uploaded).
By default, Autotag is optimized for higher precision and lower recall. You can tune the precision/recall tradeoff of your Autotag by adjusting the score threshold in your query.
It takes a minute or two to apply the Autotag to the full dataset after it is created, so if you don’t see your Autotag name suggested in the query bar immediately after you create it, try waiting a minute and then refreshing.
autotag. | Prefix for Autotag queries |
autotag.image | Prefix for image Autotag queries. This can also be used directly to match image Autotag names at the default score threshold of 0, like autotag.image = [tag_name] . |
autotag.object | Prefix for image Autotag queries. This can also be used directly to match object Autotag names at the default score threshold of 0, like autotag.object = [tag_name] . |
autotag.image.[tag_name] | The Autotag score for the image, which measures ranges from -1 to 1. |
autotag.object.[tag_name] | The Autotag score for the object, which measures ranges from -1 to 1. |
Updated over 2 years ago