Many bloggers mentioned that they want to train on the basis of the previous one, and add the pre-trained model to the last sentence of training, usually
./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg scripts/darknet53.conv.74
The last darknet53.conv.74 is the pre-trained model downloaded.
But when we train our own model, the obtained training file is a .weights file, generally yolov3_last.weights. If you want to train on this basis, you ca n’t directly use this file as a pre-trained model. If you put this directly in the training statement At the end, the training will report an error
To run the following sentence, do a fine-tuning, change yolov3_last.weights to yolov3_last.conv.23
./darknet partial cfg/darknet19_448.cfg yolov3_last.weights yolov3_last.conv.23 23
Then run the training again: ./darknet detector train cfg / voc.data cfg / yolov3-voc.cfg yolov3_last.conv.23, so that no errors will be reported