How mixed-precision training prevented me winning a competition

Introduction I usually start each new solution by creating a solid baseline from scratch, without looking to provided baselines. To my beliefs, it helps to build a deep understanding of the problem. Although my experience in competitive machine learning is quite extensive, monocular geo-pose estimation from satellite imagery was new to me. After doing some initial EDA, it became clear that data presents a few challenges to participants: * Speckle noise in the AGL map (Presumably due to way of data acquisition) * Temporal inconsistency between measured AGL values and RGB images (A good example is a fence that is visible in AGL, but missing in RGB domain, different constructions sites) * Possible data leak due to overlapping/nearby tiles from same location belonging to train/test.

xView2: Getting a medal and publishing follow-up paper on NeurIPS

The main objective of xView 2 challenge was to build a solution, capable of processing a pair of satellite images of same region (called hereinafter “pre” and “post” images) to detect buildings and classify level of damage dealt to them.

In this post I describe my solution to this problem and explain few know-how ideas, that resulted in a top-performing solution.

How to write a good code

This article is a quintessence of my all experience I’ve got for years working as a computer vision consultant. I hope you will find this interesting and useful. My goal was to create set of rules I follow personally on daily basis.

How to debug node.js addons in Visual Studio

While working on CloudCV I encountered problems in node.js addon written in native code. For CloudCV I use node.js with C++ Addon to separate high-performance algorithms (C++) from high-level networking API which node provides.

In this tutorial I’m going to reveal best practices on debugging C++ Addons for Node.js (0.12) using Visual Studio 2013.

Continue reading if you want to read in details why this works.