dun933 / tf2-yolo-nano Goto Github PK

This is a YoloNano implatement in tensorflow 2.1.0.

1.Use original yolo train set like,the train images(xxx.jpg) and the label files(xxx.txt) are in the same dir.

The label format is as follow:

classes_id x_center y_center width height ##box1 range[0.0-1.0]
classes_id x_center y_center width height ##box2 range[0.0-1.0]

Pay attention that the box data should be Normalized.

2.Create a train.txt which contains your all train images with full path,such like:

/home/user/coco/train/1.jpg
/home/user/coco/train/2.jpg
/home/user/coco/train/3.jpg

Pay attention that the picture name should correspond to the label name.

3.Replace train_path in train.py with your own train.txt path and modify the anchors with your own data.

4.Runpython3 train.py.

That's all,so easy isn't it?

1.Modify the lines below in test.py

anchors = np.array([your anchors data,shape(9,2)],dtype='float32')

test_model.load_weights('your model.h5 here')
img = cv2.imread('your test image here')

2.Run python3 test.py