If I’m not mistaken, SGDR wasn’t actually Leslie Smith’s paper. While the SGDR paper discusses a technique that’s similar to CLR, the paper mentions that Leslie Smith’s paper on CLR is “closely-related to our approach in its spirit and formulation but does not focus on restarts”.
Oh, ok. Well then, Lin Chi, welcome to Medium!
I’m fairly new to the platform myself (at least, compared to the people who have been here for 3+ years), but in my limited experience, it’s been a great place to share content.
Lin Chi, in my article, I mentioned that we generally use the (x,y) cooridinates of the center of the bounding box, the height, and the width as the 4 features to learn.
Using the center, height, and width equates to the same number of features as learning a corner, height, and width. So the amount of data required in both…
The table which I generated was completely random, so take the numbers and examples that I’ve given with a grain of salt.
You don’t convert the table into W_u and W_m. In fact, it goes the other way around. We are looking to generate (or “learn” in deep learning talk) two matrices, called W_u and W_m, such that the product…