mistasse's

The concept of Non-Maximum Suppression (NMS) is pervasive in Deep Learning solutions, yet I have found it to be rarely discussed, and even less formalized. This is an attempt at demystifying and formalizing it as well as I can. This is an undertaking I have already have started in my thesis, Section 2.2.5, but now is an opportunity to make it more accessible. Basically, NMS is useful when we want to obtain a sparse set of outputs from a model architecture that, for one or more reasons, does not do it naturally.

Lately, I’ve learned about PyScript and Pyodide. Pyodide provides a CPython port to WASM, i.e. capable of running in the browser, and PyScript makes it simple to integrate Python scripts in a webpage (as simple as JavaScript), manipulate the DOM, use JS libraries, …