Prerquisites

Blender - tested with 2.79 - www.blender.org
VOTT - https://github.com/Microsoft/VoTT
Download actual build from https://github.com/Microsoft/VoTT for a Visual Tagging Tool
pip install azure-cognitiveservices-vision-customvision

Blender Activties

Add Stl to empty Blender or using py Function CreateFromScratch(pathToStl,"c:\temp\render.blend") to create new Scene

(
## Set Active Camera
Move Active Camera to View
Hotkey: Ctrl-Alt-Numpad0
# Check with
Hotkey: Numpad0
)

#Run Script
path "c:\Program Files\Blender Foundation\Blender\2.79"
#Run via blender to ease the use of bpy module
blender --background -P render.py

Process

After the Process you receive a json file with metadata including Boundingboxes and the rendered pictures

Object Detection Yolo

USE Azure DataScienceVM to train the model:

Setup GPU Machine for Darknet Training

Install Cuda on 18.04

wget https://developer.nvidia.com/compute/cuda/9.2/Prod2/local_installers/cuda_9.2.148_396.37_linux
wget https://developer.nvidia.com/compute/cuda/9.2/Prod2/patches/1/cuda_9.2.148.1_linux
sudo apt install -y make gcc freeglut3 freeglut3-dev libxi-dev libxmu-dev
Sudo sh cuda_9.2.148_396.37_linux

Follow the instructions
- accept Eula
- agree to non unspported configuration
- agree to NVIDIA Accelerated Graphics Driver
- agree to OPENGL Driver
- disagree to nvidia-xconfig
- agree to CUDA 9.2 Toolkit (use default for the subsequent)
- agree to install the Samples

Install patch*
Sudo sh cuda_9.2.148.1_linux

''Testing Installation**
Change to deviceQuery Folder:
cd samples/1_Utilities/deviceQuery/
make
Query for existing CUDA enabled Cards:
./devicequery

Install Darknet and Compile for GPU

git clone https://github.com/pjreddie/darknet
edit Makefile
and change GPU=0 to GPU=1 (DEGUG=0 to DEBUG=1)
make

test installation

./darknet detect cfg/yolov3.cfg yolov3.weights data/dog.jpg

#Train the model [!TODO]

Build and Test

Custom Vision

Upload to CustomVision

./python VOTTtoCV.py

Export to customvision

Score with Customvision