CNN for Semantic Segmentation
Explore how to use a pretrained DeepLabV3 convolutional neural network for semantic segmentation to identify and map object shapes in images. Learn to generate pixel-level classifications that enable precise measurements of object positions and orientations for automated inspection tasks.
We'll cover the following...
When we discussed CNN for Object Detection, we obtained a rough location for the objects of interest through a bounding box. That is enough in some circumstances, but sometimes we need to know the shape of objects, in addition to their position.
Consider the image below:
Assume that our automated inspection task requires measuring the (x, y) coordinates of some critical microcontroller components in the left-hand side image, along with their orientation in the board plane. We would need a mask for each component such that their blob centroid and main axis angle can be computed.
Semantic segmentation is the computation of a multi-class map from an image, giving the surface of each object of interest. ...