Summary of AI Learning Catalog
1. Brief introduction
Through the previous study, the principle and implementation of the image classification model have been understood. Image classification assumes that there is only one target in the image, and the algorithm is to classify the entire image.
Let's learn "target detection", that is, to find the desired target from an image and mark the position.
2. Bounding box
Bounding box: The bounding box is a box, using the box to frame the target just right.
There are two ways to express the position and size of the bounding box with coordinates
- Coordinate representation of the upper left corner and lower right corner of the bounding box
- The coordinates of the center of the bounding box and the width and height of the box.
3. Anchor frame
Anchor box: anchor box, imagine, in fact, we do not know the size and position of the target, first assume the center position of the target, similar to the anchor of a ship, and then use multiple frames of different sizes to detect whether there is a target to be measured.
Boxes of different sizes can be represented by different scaling and aspect ratios.
In addition to using coordinates to describe the size and position of the anchor box, a label is also needed to mark the category of the detected object.
4. Cross-merge ratio
Intersection ratio: IoU, used to measure the similarity between the anchor box and the real bounding box, the Jaccard coefficient, the mathematical formula is as follows: