I am in Course 4, week 3 where Andrew Ng explains how it is possible to vectorise the sliding window algorithm computation for object detection. Why has he not explained what happens to the algorithm when:
For instance, if 2 or more sliding windows detect the same car in an image because the size of the car can be bigger than the size of the sliding window? How will the object be detected and be bounded accurately with a box? What if there are two cars in one sliding window?
I know I will probably find an answer to this by the time I am done with the exercises. But I feel uncomfortable watching all the explanations without an answer to my question.
You will find that Prof Andrew gives a beautiful dedicated explanation for this in the lecture video titled “Non-max Suppression”. If you want, you can watch this lecture video first before watching the others, and your query will be resolved.
Now, this part is beautifully explained in the video titled “Anchor Boxes”. I can try and explain you the answers in this post, but with all due respect, I can’t explain nowhere close to the kind Prof Andrew can, so, once again, you can watch these lecture videos first, and then go back and watch the remaining videos. Let me know if this helps.