Hi, after finishing week 3 I was wondering what the downside of the U-Net is. It seems to have less computational cost than yolo and isn’t it always better to classify every pixel. If we want a box we could also draw from the pixel borders right?
Like a lot of things in engineering I think the answer depends on the requirements/ business objectives. Take this segmented image, for example
Most likely all those cars on the right side have the same class identifier and all of the pixels should be the same color. Maybe all you need to know is that there is a huge blob of car pixels. But if you need to know how many cars are near by, and where each one begins and ends, I’m not sure pixel level segmentation is the most appropriate solution.
1 Like
Thank you, good point.