More a sugestion than an issue

[JoseRicardoGomes]

Hello,

I've trying on and off for the sake of my sanity to get YOLO and yolo_mark to work on my machine for a final year college project.
It's been about 12 weeks since I started and, somehow, didn't manage to get anything to work, the instructions you provide are, for the most part, a bit unclear and shallow and every time something seems to be broken. It's either that or I'm just too dumb to understand and use this.

I would suggest for you to pair up with someone willing to go through and write the documentation in a better, more understandable and robust format.

Nevertheless this is some aw standing work and you deserve my admiration for that.

Thank you,

JRG

[AlexeyAB]

Is your problem in compilation, training or detection?

[JoseRicardoGomes]

Had problems in understanding the instructions for the three of them.
Understanding what the auxiliary files you require do and must be located and the whole workflow of both yolo and yolo_mark It's not clear.

[AlexeyAB]

• Do you mean that you could not compile darknet in 12 weeks?
• Can you give some good instructions on another machine learning framework and object-detection neural network that you managed to build and train quickly?
• What OS do you use?
• Did you successfully install OpenCV, CUDA, cuDNN?
• What is your primary programming language Python, C, ...?

[JoseRicardoGomes]

Yes I do. Elsewise I would not humiliate myself like this.

I also tried Keras and none of anything seems to work. Implement (or build as we say now) as instructed in one place booom, something is broken, try another one, solves some issue that the previous elevated but other 3 emerge, try de official, instructions are fussy, incomplete or ambiguous.
I've not been able to do anything but trying to find fixes and troubleshooting for the past 12 weeks (hence I think the problem lies within me).

Please bare with me here, I come from a judiciary police/forensics background, things I use are usually very succinct and leave no room for guessing.

I use Ubuntu 18.04.4 LTS
12 GiB of RAM
intel core i7-4510U dual core, multi threaded processor
amd graphics card, don't bother, never worked on any distribuition.

opencv version: 4.3.0-dev as per my python interpreter
both python2 and 3 fully supported by the system.

Don't have a primary but I grasp C, C++, python, R, java and C# (also some arcane javascript).

Sorry if I sound angry and frustrated, I am and I'm not blaming you.

[AlexeyAB]

• Did you read requirements? https://github.com/AlexeyAB/darknet#requirements

  • It can't use AMD graphic card, since it requires nVidia GPU https://en.wikipedia.org/wiki/CUDA#GPUs_supported
  • It requires OpenCV installation, run sudo apt-get install libopencv-dev

• Did you try make to build Darknet?

The simplest way:

1. Download and un-zip Darket https://github.com/AlexeyAB/darknet/archive/master.zip
2. run make in the Darknet folder
3. Download file https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights to the Darknet folder
4. Run in the darknet folder:
   ./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights data/dog.jpg

I try to understand which part is the most incomprehensible and difficult.
Also, if you try to train many other object detectors from other frameworks, then yolo will seem like a paradise to you)
In general, all machine learning frameworks are designed for developers, not for users.

[JoseRicardoGomes]

Will try that.
My major is computer engineering btw.

Get back to you as soon as I can.

Thank you for your time.

[AlexeyAB]

https://www.youtube.com/watch?v=mKAEGSxwOAY