Can someone explain how dimensions of the anchor boxes are calculated from anchor ANCHOR_SCALES and ANCHOR_RATIOS? How do they relate to generating 1:1, 1:2 or 2:1 aspect ratio anchor boxes with box areas 128^2, 256^2 as mentioned in the Faster RCNN paper?

Sorry to bother you.

When I delete the lib.utils. ,the test.py is right but the demo.py is wrong , however,when I use the lib.utils. , the demo.py is right but the test.py is wrong !!

why? why

The 2x down-sampling is one of the important operations in reference models. But, a convolution or a pooling with stride=2, padding='SAME' may result in different outputs over different deep learning libraries (e.g., TensorFlow, CNTK, Theano, Caffe, Torch, ...) due to their different padding behaviors.

For example (TensorNets syntax; but can be regarded as pseudo codes for other libraries),

def intersection_area(rect1,rect2):
r1 = cv2.rotatedRectangleIntersection(rect1, rect2)
if r1[1] is None:
return 0
# order_points=cv2.convexHull(r1[1],returnPoints=True)
area = cv2.contourArea(r1[1])
return area
def intersection_area_convex(rect1,rect2):
r1 = cv2.rotatedRectangleIntersection(rect1, rect2)
if r1[1] is None:
return 0
order_p

Mac OSX 10.14.1 (18B75)

Traceback (most recent call last):
File "yolo_opencv.py", line 98, in
draw_prediction(image, class_ids[i], confidences[i], round(x), round(y), round(x+w), round(y+h))
File "yolo_opencv.py", line 39, in draw_prediction
cv2.rectangle(img, (x,y), (x_plus_w,y_plus_h), color, 2)
TypeError: only length-1 arrays can be converted to Python scalars