Finally got results, all you need is a file with each class per line (classes.txt), the configuration (yolov3-train.cfg) and the weights for the modified YOLO detection (in this case I used older ones insted of the final ones so yolov3-train_last.weights).
This gave me a result, as bad as expedcted, but in the right area at least. Imagine if my machine could do it itself.
You can find the files in my GitHub repository.