-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathmask-ensemble-5fold.py
138 lines (114 loc) · 4.85 KB
/
mask-ensemble-5fold.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
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
# @Author: Xuan Cao <xuan>
# @Date: 2019-12-22, 1:25:27
# @Last modified by: xuan
# @Last modified time: 2019-12-22, 1:33:27
import warnings
warnings.filterwarnings("ignore")
from pathlib import Path
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import tqdm
from util.utils import *
def compute_detail_score(df, dice):
res = []
res.append(df[dice].mean())
#c1 -> c4 dice
for label in ['Fish', 'Flower', 'Gravel', 'Sugar']:
df_tmp = df[df['cls'] == label]
res.append(df_tmp[dice].mean())
# neg & pos dice
res.append(df[df['truth'] == ''][dice].mean())
res.append(df[df['truth'] != ''][dice].mean())
# c1 -> c4 pos
for label in ['Fish', 'Flower', 'Gravel', 'Sugar']:
df_tmp = df[df['cls'] == label]
res.append(df_tmp[df_tmp['truth'] != ''][dice].mean())
return res
def ensemble_rles(rles1, rles2, mode='intersect'):
res = []
for rle1, rle2 in tqdm.tqdm(zip(rles1, rles2)):
m1 = rle2mask(rle1, height=350, width=525, fill_value=1)
m2 = rle2mask(rle2, height=350, width=525, fill_value=1)
if mode == 'intersect':
mask = ((m1+m2) == 2).astype(int)
elif mode == 'union':
mask = ((m1+m2) > 0).astype(int)
else:
RuntimeError('%s not implemented.'%mode)
rle = mask2rle(mask)
res.append(rle)
return res
def ensemble_rles_multi(rles, ts=0):
res = []
for rle in tqdm.tqdm(rles):
m = np.zeros((350, 525))
for x in rle:
m += rle2mask(x, height=350, width=525, fill_value=1)
mask = (m > ts).astype(int)
ans = mask2rle(mask)
res.append(ans)
return res
def load_seg_pred(seg_name, name, tta):
#load val
df_val = []
for fold in range(5):
if tta <= 1:
df_val.append(pd.read_csv('../output/'+ seg_name + '/' + 'valid_fold%d.csv'%fold))
else:
df_val.append(pd.read_csv('../output/'+ seg_name + '/' + 'valid_fold%d_tta%d.csv'%(fold, tta)))
df_val = pd.concat(df_val)
df_test = pd.read_csv('../output/'+ seg_name + '/' + 'test_5fold_tta%d.csv'%tta)
df_val.rename(columns={'EncodedPixels': name}, inplace=True)
df_test.rename(columns={'EncodedPixels': name}, inplace=True)
return df_val, df_test
df_train = pd.read_csv('../input/train_350.csv')
df_train.rename(columns={'EncodedPixels': 'truth'}, inplace=True)
_save=1
tta=3
ts=2
ensemble_file_name = 'b5-Unet-inception-FPN-b7-Unet-b7-FPN-b7-FPNPL'
segs = [
'efficientnetb5-Unet-DICE-warmRestart-10x3-bs16', #0.66823, 0.67019
'inceptionresnetv2-FPN-DICE-warmRestart-10x3-bs16', #0.67133, 0.67311
'efficientnetb7-Unet-DICE-warmRestart-10x3-bs16', #0.67060, 0.67268
'efficientnetb7-FPN-DICE-warmRestart-10x3-bs16', #0.67070, 0.67253
'efficientnetb7-FPN-DICE-warmRestart-10x3-bs16-pl', #0.67332, 0.67511
]
# load seg results
for i in range(len(segs)):
df1, df2 = load_seg_pred(segs[i], 's%d'%(i+1), tta)
if i == 0:
df_seg_val, df_seg_test = df1, df2
else:
df_seg_val['s%d'%(i+1)] = df1['s%d'%(i+1)]
df_seg_test['s%d'%(i+1)] = df2['s%d'%(i+1)]
## merge seg valid
df_seg_val = pd.merge(df_seg_val, df_train, how='left')
df_seg_val.fillna('', inplace=True)
df_seg_val['s%d'%(len(segs)+1)] = ensemble_rles_multi(df_seg_val[['s%d'%(i+1) for i in range(len(segs))]].values.tolist(), ts)
# Compute val score
print('Compute dice score.')
df_seg_val.fillna('', inplace=True)
df_seg_val['cls'] = df_seg_val['Image_Label'].apply(lambda x: x.split('_')[-1])
for i in range(len(segs)):
df_seg_val['dice%d'%(i+1)] = df_seg_val.apply(lambda x: dice_np_rle(x['s%d'%(i+1)], x['truth']), axis=1)
df_seg_val['dice%d'%(len(segs)+1)] = df_seg_val.apply(lambda x: dice_np_rle(x['s%d'%(len(segs)+1)], x['truth']), axis=1)
if _save:
df_seg_val.rename(columns={'s%d'%(len(segs)+1): 'EncodedPixels'}, inplace=True)
df_seg_val.to_csv('../output/%s/valid_5fold_tta%d.csv'%(ensemble_file_name, tta), index=None)
print('')
print(' dice | c1 | c2 | c3 | c4 | neg | pos | pos1 | pos2 | pos3 | pos4')
for i in range(len(segs)):
res = compute_detail_score(df_seg_val, 'dice%d'%(i+1))
res = ','.join('%.5f'%x for x in res)
print('Seg-%d scores: %s'%(i+1, res))
res = compute_detail_score(df_seg_val, 'dice%d'%(len(segs)+1))
res = ','.join('%.5f'%x for x in res)
print('ensemble scores: %s'%res)
if _save:
df_seg_test.fillna('', inplace=True)
df_seg_test['EncodedPixels'] = ensemble_rles_multi(df_seg_test[['s%d'%(i+1) for i in range(len(segs))]].values.tolist(), ts)
df_seg_test.to_csv('../output/%s/test_5fold_tta%d_details.csv'%(ensemble_file_name, tta), index=None)
df_seg_test.to_csv('../output/%s/test_5fold_tta%d.csv'%(ensemble_file_name, tta), index=None, columns=['Image_Label', 'EncodedPixels'])