Darknet training script doesn't find setup files

[ghost]

I followed the tutorial step by step on how to train YOLOv4 on custom data (where files should be put and what data should be provided). There were some conflicts such as :

• Step 3 says "Create file obj.data in the directory build\darknet\x64\data\, containing (where classes = number of objects)". However, when executing the train script the file was referenced in data\obj.data
• Step 4 says "Create file train.txt in directory build\darknet\x64\data\, with filenames of your images, each filename in new line, with path relative to darknet.exe". Again, the path is ambiguous.

Anyway, I tried running the script (./darknet detector train ...) without changing anything and the script couldn't find train.txt and test.txt files (which wasn't a surprise). I tried changing around the paths to fit accordingly but my it still doesn't find my setup files.

Learning Rate: 0.001, Momentum: 0.949, Decay: 0.0005
 Detection layer: 139 - type = 28 
 Detection layer: 150 - type = 28 
 Detection layer: 161 - type = 28 
Couldn't open file: data/train.txt

I scrolled through the issues and no one seems to have had it before. So there's high chance I'm missing sth obvious. However, any help would be appreciated. Thank you.

[stephanecharette]

If you are on Linux, you may want to try using DarkMark to create all of the files. It is a GUI, and it sets everything up so there is no problem.

To remove any ambiguity, I always use the full absolute paths for all filenames in the config files. Note that the darknet executable doesn't have to be run from the darknet directory. Everything should be self-contained within that executable. I typically setup my projects like this:

~/darknet/cfg/...
~/darknet/darknet
...
~/nn/foo/
~/nn/foo/foo.cfg
~/nn/foo/foo.names
~/nn/foo/set_01/image_001.jpg
~/nn/foo/set_01/image_002.jpg ...etc...

Then to train:

cd ~/nn/foo/
~/darknet/darknet detector -map -dont_show train foo.data foo.cfg 2>&1 | tee --append output.log

[ghost]

Thank you for your answer. I am working on google colab and my dataset is already labeled. But thank you for the help on the setup !