-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate.py
48 lines (36 loc) · 1.54 KB
/
evaluate.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
# -*- coding: utf-8 -*-
import numpy as np
# ----------- this 0.07 is unchangeable ------#
ERROR_RANGE = 0.07
class Evaluator:
# 17.11.19.1
boundary_offset = 15
def get_boundary_mask(self, array, ignore_boundary=True):
assert array.ndim >= 2
hei, wid = array.shape[0], array.shape[1]
if ignore_boundary:
mask = np.full(array.shape, fill_value=0, dtype=np.bool)
f_offset = self.boundary_offset
mask[f_offset:hei-f_offset, f_offset:wid-f_offset] = True
else:
mask = np.full(array.shape, fill_value=1, dtype=np.bool)
return mask
def error_acc(self, disp_pre, disp_gt, error_range=ERROR_RANGE, ignore_boundary=True, eval_mask=None):
err_pre = (abs(disp_pre - disp_gt) <= error_range).astype(np.uint8)
#acc_no_mask = np.mean(err_pre)
mask = self.get_boundary_mask(disp_gt, ignore_boundary)
if eval_mask is not None:
mask *= eval_mask
valid_err_pre = err_pre[mask]
value_acc = np.mean(valid_err_pre)
# boundary as unvalid
err_pre[~mask] = 1
return err_pre, value_acc #, acc_no_mask
def mse(self, algo_result, disp_gt, factor=100, ignore_boundary=True, eval_mask=None):
mask = self.get_boundary_mask(disp_gt, ignore_boundary)
if eval_mask is not None:
mask *= eval_mask
with np.errstate(invalid="ignore"):
diff = np.square(disp_gt - algo_result)
diff[~mask] = 0
return diff, np.average(diff[mask]) * factor