Creating training material for YoloV3

1st Alternative: YOLO-Annotation-Tool

We went to a Pool & Snooker Bar called Corona and got some footage for our project.

Next we used YOLO-Annotation-Tool to create training sets for YOLO.

git clone

cd YOLO-Annotation-Tool

Move our images to the 001 directory under ./YOLO-Annotation-Tool/Images .

mv ./SnookerData/*.jpeg ./YOLO-Annotation-Tool/Images/001/

We need to remove the cat photos that are in the 001 directory which are all .jpg files.

cd ./YOLO-Annotation-Tool/Images/001
rm *.jpg

Next convert our .JPEG files to .JPG files

mogrify -format jpg *.jpeg

To be able to run we needed a few packages from apt.

sudo apt-get install python-tk python-pil python-imaging-tk
sudo pip install Image

Now we should be able to run


The Labeling-Tool looks like this:

Alt Text

Although the labeling works well, we wouldn’t get the program to run or succesfully. It would either create empty files or say that we didn’t have some obscure directories (e.g. a directory with the name of our image).

2nd Alternative: Open Labeling

We also tried another labeling-tool called Open Labeling.

git clone

You can install everything at once by simply running:

python -mpip install -U pip
python -mpip install -U -r requirements.txt

Ran the program, shut it down and tried reopening it again and was greeted by Error messages. That was the first and last time we got it to work.

3rd Alternative: Yolo_mark by AlexeyAB

Luckily we found an Annotation Tool called Yolo_mark by the creator of Darknet, AlexeyAB.

First off:

git clone
cd Yolo_mark

To compile it we ran 3 commands:

cmake .
bash # Ctrl + C Ends

Set the number of classes (objects) in /x64/Release/yolo-obj.cfg on line 230. Set filter value in /x64/Release/yolo-obj.cfg on line 224.

  • for YoloV2 (classes + 5)*3

Now run Yolo_mark again and start making your BBoxes.

Alt Text

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.