xview2 Challenge

MAIN GOAL / DELIVERABLES

After spending some time studying and doing some easier projects on Deep Learning, I would like to test my knowledge.

For that reason, I decided to spend this month (December 2019) participating in this challenge.

I will share my analysis, models and evolution step-by-step.

STEP 1. REPLICATE BASELINE

Using VSCode with Jupyter notebook

Remote SSH to VM using VSCode

Create project, set configs and create virtual environment

Obs.: Used anaconda and python 3.6 to replicate xview2-baseline

Download the data and upload to Google Cloude Storage

To download the data (28GB) directly to my virtual machine, I needed to access their website via browser. This medium post "Your desktop on Google Cloud Platform" showed me how to, step by stem:

Create VM
Install a VNC server
Install desktop environment
Set up VNC server
Install VNC client
Open the firewall
Connect to the VNC server

STEP 2. Create TFRecods

Localization

STEP 3. OPTIMIZE PIPELINE WITH TFRecords AND DISTRIBUTED TRAINING

STEP 4. WHAT NEXT?

SOME PROBLEMS THAT I ENCOUNTERED DURING THIS EXPERIMENT:

NEXT STEPS:

REFERENCES:

About

Computer Vision for Building Damage Assessment using satellite imagery of natural disasters

Languages

Language:Jupyter Notebook 95.9%Language:Python 3.3%Language:Makefile 0.6%Language:Shell 0.2%