Object Detector

Object detectors find objects in an image by computing regions of interest (ROI), then draw a bounding box around it, and finally, run a classification on the ROI to predict the object's class.

The output of an object detector is a set of bounding boxes including its (x,y) coordinates, width, height, and the assigned class with a confidence score. Its performance is usually measured with the mAP.

In comparison to semantic segmentors, object detectors can identify specific objects and count the occurrences in an image. However, unless you're labeling rectangular objects, the bounding boxes will contain some noise influencing the classifier's performance negatively—instance segmentation yields better results often. In comparison to instance segmentors, object detectors require less computational resources and less labeling effort, though.

Do you have annotated data with bounding boxes for object detection but don't get good performance? You could try out instance segmentation as it usually performs better. Hasty.ai offers features to almost automatically convert your bounding boxes to masks.

Further resources

Last updated on Jun 01, 2022

Removing the risk from vision AI.

Only 13% of vision AI projects make it to production, with Hasty we boost that number to 100%.