Object detection finds what is in an image and where it is.
The output is a bounding box plus a class label plus a confidence score.
Two-stage detectors like Faster R-CNN propose regions and then classify them.
One-stage detectors predict boxes and classes in a single pass.
Intersection over Union measures how well a predicted box matches the true one.
Classification says what is in an image, and detection says what and where.
The output is a list of boxes, each with a class and a confidence score.
That is harder than it sounds, since the number of objects is not known in advance.
An image might hold nothing or forty things, and the model must handle both.
Two-stage detectors split the job in half.
A region proposal network suggests places worth examining.
A classifier then decides what each proposed region contains.
Faster R-CNN is the standard example and is accurate and comparatively slow.
One-stage detectors predict boxes and classes together in a single pass.
YOLO and SSD are the standard examples, and they are fast enough for video.
The gap in accuracy between the families has narrowed considerably.
Every detector produces many overlapping boxes for the same object.
Non-maximum suppression keeps the most confident box and discards those overlapping it heavily.
Intersection over Union measures that overlap and decides what counts as a correct box.
Mean average precision is the usual headline metric, computed across classes and thresholds.
One question on this card. Every option explains itself once you pick, right or wrong.
22 concepts in this module, then a graded test with every answer explained.