AI Image Classification tools – drawing boxes for YOLOv4 is a pain but there is help

Posted on by

I am classifying using YOLOv4 and using AlexeyAB’s version of Darknet. He also has a classifying tool (putting boxes on images) for it called Yolo_mark. I signed up for a course and the teacher (who is totally helping me out – super guy on the video and in real life) suggests Supervisor.ly but it seems to be an online tool and I have thousands of images to mark/classify (oh and I am from Dutch descent – so the enterprise version will not be an option as there is no downloadable trial version). So his course isn’t quite ready (he said a couple weeks) where he will demo this so I have searched a bit and I found a VERY basic tool that is keyboard driven except selecting and moving the box. SUPER FAST TO USE it turns out. It needed a couple tweaks too so here goes ( I should mention I am following this guy’s guide when AlexeyAB’s instructions got a little vague) …

It says to download it, then open up the .sln in VisualStudio. I already have the compiler downloaded (forget the name dangit). Anyways – there are these user variables like $(MY_OWN_VAR) that you can use like “environment variables withing VS” but he had EVERYTHING perfectly setup to use those variables – but how do I set them? Well a file was missing where these would be stored. So these variables are (stupidly) called “macros” . Finding this phraseology too 15 minutes on Google – once I found out what they were called it was a snap.

What is missing to get these macros “set” is a file – it is called User Macros – or a settings file. Here is a demo that if you follow will make it come alive. Then – you have to add that file as a file to all the compile environments that you have – so in my case Debug – x64, Release x64. The main point is to go to View->Property Manager->Create New File. Once it is created – add the macro – in this case the one I am focused on was called

  • OPENCV_DIR and set it to C:\opencv\build\install
  • Then add this macros file to all the builds (just in case really) in the Project Manager tab

Then it compiled just like he said it would.

Couple more steps:

  • When I moved the whole shebang to another computer – I found I had to install Microsoft’s re-distributable 2015-2019
  • Also had to move a file from OpenCV directory to the main directory (called Release) called opencv_world410.dll. I had to do a search using Windows Explorer to find it in OpenCV (in my case c:\opencv). This is likely because the OpenCV directory was not in my path.

All works! Super easy tool to use. It is so easy it is frustrating at first til you realize that life COULD BE SIMPLER . Thanks again for the tool

He prof’s favourite tools for this seem to be Supervisor.ly, Colab Labeller, Dark Label, LabelImg, LabelBox. So far, Yolo_mark is mine (cause I have only tried LabelImg for Pascal VOC format). See my article of what I used to transform from Pascal VOC to YOLOv4 format.