|
|
|
@ -166,7 +166,7 @@ def _preprocess_true_boxes(true_boxes, anchors, in_shape, num_classes,
|
|
|
|
|
valid_mask = boxes_wh[..., 0] > 0
|
|
|
|
|
|
|
|
|
|
wh = boxes_wh[valid_mask]
|
|
|
|
|
if wh:
|
|
|
|
|
if wh.size > 0:
|
|
|
|
|
wh = np.expand_dims(wh, -2)
|
|
|
|
|
boxes_max = wh / 2.
|
|
|
|
|
boxes_min = -boxes_max
|
|
|
|
@ -319,7 +319,7 @@ def _choose_candidate_by_constraints(max_trial, input_w, input_h, image_w, image
|
|
|
|
|
dx = int(_rand(0, input_w - nw))
|
|
|
|
|
dy = int(_rand(0, input_h - nh))
|
|
|
|
|
|
|
|
|
|
if box:
|
|
|
|
|
if box.size > 0:
|
|
|
|
|
t_box = copy.deepcopy(box)
|
|
|
|
|
t_box[:, [0, 2]] = t_box[:, [0, 2]] * float(nw) / float(image_w) + dx
|
|
|
|
|
t_box[:, [1, 3]] = t_box[:, [1, 3]] * float(nh) / float(image_h) + dy
|
|
|
|
|