def eval_single_frame(target, box):
"""
input params:
target, python ordered dict
box, sorted boxes dict from predictions
"""
TP = 0
FP = 0
FN = 0
precision = 0
recall = 0
F_measure = 0
if not len(box['text_lines']) == 0:
for t in target:
d = np.array(t, dtype='int32')
is_best = 0
for m in box['text_lines']:
n = np.array([m['x0'], m['y0'], m['x1'], m['y1'], m['x2'],
m['y2'], m['x3'], m['y3']], dtype='int32')
# pick out the best match
iou = intersection(n, d)
if iou>is_best:
is_best = iou
if is_best > 0.5:
TP = TP+1
elif is_best == 0:
FN = FN +1
else:
FP = FP+1
if TP > 0:
precision = TP/(TP+FP)
recall = TP/(TP+FN)
F_measure = 2*precision*recall/(precision+recall)
return precision, recall, F_measure
def eval_single(bb_gt, bb_pred):
"""
input params:
bb_gt, numpy array, (batch_size, time_steps, 40*9)
bb_pred, ((batch_size, time_steps, 40*9))
"""
length_vect = bb_gt.shape[2]
unit = 9
num = int(length_vect/unit)
is_best = 0
TP = 0
FP = 0
FN = 0
for i in range(num):
for j in range(num):
# pick out the best match
iou = intersection(bb_gt[:,:,i*unit:i*unit+8], bb_pred[:,:,i*unit:i*unit+8])
if iou>is_best:
is_best = iou
if iou > 0.5:
TP = TP+1
elif iou > 0:
FP = FP+1
else:
FN = FN+1
precision = TP/(TP+FP)
recall = TP/(TP+FN)
F_measure = 2*precision*recall/(precision+recall)
return precision, recall, F_measure