TUTORIAL.md

OpenCV-Minecraft Tutorial

Capture a video

First you need to capture an video on Minecraft

I've personnaly capture a 2 minutes long video of me flying and walking over a world in minecraft without any trees

Now you need to install cv2 with pip3 install opencv-python and matplotlib with pip3 install matplotlib

Convert to multiple images

After that, you need to convert the mp4 video into multiple image with ffmpeg (you can install it using pip3 install ffmpeg-python)

ffmpeg -i input.mp4 -qscale:v 2 -vf fps=3 ./negativeImage/out%d.jpg

• -vf fps=3 is to choose how much image do you want every frame, this one output 3 images every seconds
• -qscale:v 2 is to choose the quality (2 is excellent quality and 31 is the worst quality)

You need to put this images into the ./negativeImage folder.

Now go the the root of the folder and use this command to create a file that contain all the references of the images

cd negativeImage
ls *.jpg > negatives.txt

You now need to install the libopencv-dev library using apt-get install libopencv-dev to continue

opencv_createsamples -img trees/1.jpg -bg negativeImage/negatives.txt -info sampleImageTest/cropped1.txt -num 128 -maxxangle 0.0 -maxyangle 0.0 -maxzangle 0.3 -bgcolor 255 -bgthresh 8 -w 48 -h 48

Next, you need to collect all the description files and combine into one file

cat sampleImageTest/cropped*.txt > sampleImageTest/positives.txt

Then combine all the images into a vec file

opencv_createsamples -info sampleImageTest/positives.txt -bg negativeImageDirectory/negatives.txt -vec cropped.vec -num 250 -w 48 -h 48

• -num 250 is the number of positives images you have

And finally train our Haar classifier with the following command:

cd negativeImage 
opencv_traincascade -data ../classifier -vec ../cropped.vec -bg negatives.txt -numPos 200 -numNeg 600 -numStages 10 -precalcValBufSize 1024 -precalcIdxBufSize 1024 -featureType HAAR -minHitRate 0.995 -maxFalseAlarmRate 0.5 -w 48 -h 48

• -numPos 200 is the number of positives images you have (with a margin because opencv can take more positive images that you have some times)
• -numNeg 600 is the number of negatives images

Final result with Google Colab