How to train If I add some pictures?

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.75k stars 7.96k forks source link

How to train If I add some pictures? #1209

Open mahxn0 opened 6 years ago

mahxn0 commented 6 years ago

First, I trained a weights not detect well Then, I add some pictures to train_img, can I train it after first trained model? Is there some defferent train with darknet53.conv?

after

./darknet detector train cfg/coco.data cfg/yolov3.cfg darknet53.conv

6000 pictures , 40000 iters I get a model yolo_final.weights not take good. Then I add 500 negative pictures to train_img , can i fine tuning from first result yolov3_final.weights? not start from darknet53.conv ...

AlexeyAB commented 6 years ago

@mahxn0 Yes, you can continue training from yolov3_40000.weights instead of darknet53.conv, if there wasn't changed number of classes.

mahxn0 commented 6 years ago

thank you, If continue, train.list is all of my picture include 500 negative pic? how to set iter num? like 45000? 50000? or you mean only used yolo40000.weights as pretrain model ?

AlexeyAB commented 6 years ago

train.txt should contain old+new images.
Just run ./darknet detector train cfg/coco.data cfg/yolov3.cfg backup/yolov3_40000.weights

You can increase steps=50000,60000 and max_batches=70000

Diaislam commented 2 years ago

@AlexeyAB this will take forever is there anyway to train faster especially when adding images every day?