-
Notifications
You must be signed in to change notification settings - Fork 2
/
eval_window.py
170 lines (136 loc) · 6.29 KB
/
eval_window.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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
import ever as er
from ever.core.builder import make_model, make_dataloader
import torch
import numpy as np
import os
from data.loveda import COLOR_MAP
import logging
from ever.core.checkpoint import load_model_state_dict_from_ckpt
from ever.core.config import import_config
from train_loveda import seed_torch
import argparse
from albumentations import Compose, Normalize
# from ever.magic.bigimage import sliding_window
from tqdm import tqdm
import math
from torch.nn.modules.utils import _pair
seed_torch(2333)
logger = logging.getLogger(__name__)
er.registry.register_all()
def sliding_window(input_size, kernel_size, stride):
ih, iw = input_size
kh, kw = _pair(kernel_size)
sh, sw = _pair(stride)
assert ih > 0 and iw > 0 and kh > 0 and kw > 0 and sh > 0 and sw > 0
kh = ih if kh > ih else kh
kw = iw if kw > iw else kw
num_rows = math.ceil((ih - kh) / sh) if math.ceil((ih - kh) / sh) * sh + kh >= ih else math.ceil(
(ih - kh) / sh) + 1
num_cols = math.ceil((iw - kw) / sw) if math.ceil((iw - kw) / sw) * sw + kw >= iw else math.ceil(
(iw - kw) / sw) + 1
x, y = np.meshgrid(np.arange(num_cols + 1), np.arange(num_rows + 1))
xmin = x * sw
ymin = y * sh
xmin = xmin.ravel()
ymin = ymin.ravel()
xmin_offset = np.where(xmin + kw > iw, iw - xmin - kw, np.zeros_like(xmin))
ymin_offset = np.where(ymin + kh > ih, ih - ymin - kh, np.zeros_like(ymin))
boxes = np.stack([xmin + xmin_offset, ymin + ymin_offset,
np.minimum(xmin + kw, iw), np.minimum(ymin + kh, ih)], axis=1)
return boxes
class SegmSlidingWinInference(object):
def __init__(self):
super(SegmSlidingWinInference, self).__init__()
self._h = None
self._w = None
self.device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
def patch(self, input_size, patch_size, stride, transforms=None):
""" divide large image into small patches.
Returns:
"""
self.wins = sliding_window(input_size, patch_size, stride)
self.transforms = transforms
return self
def merge(self, out_list):
pred_list, win_list = list(zip(*out_list))
num_classes = pred_list[0].size(1)
res_img = torch.zeros(pred_list[0].size(0), num_classes, self._h, self._w, dtype=torch.float32)
res_count = torch.zeros(self._h, self._w, dtype=torch.float32)
for pred, win in zip(pred_list, win_list):
res_count[win[1]:win[3], win[0]: win[2]] += 1
res_img[:, :, win[1]:win[3], win[0]: win[2]] += pred.cpu()
avg_res_img = res_img / res_count
return avg_res_img
def forward(self, model, image_tensor, **kwargs):
assert self.wins is not None, 'patch must be performed before forward.'
# set the image height and width
self._h, self._w = image_tensor.shape[2:4]
return self._forward(model, image_tensor, **kwargs)
def _forward(self, model, image_tensor, **kwargs):
self.device = kwargs.get('device', self.device)
size_divisor = kwargs.get('size_divisor', None)
assert self.wins is not None, 'patch must be performed before forward.'
out_list = []
for win in tqdm(self.wins):
x1, y1, x2, y2 = win
image = image_tensor[: ,:, y1:y2, x1:x2]
if self.transforms is not None:
image = self.transforms(image=image)['image']
h, w = image.shape[2:4]
if size_divisor is not None:
image = er.preprocess.function.th_divisible_pad(image, size_divisor)
image = image.to(self.device)
with torch.no_grad():
out = model(image)
if size_divisor is not None:
out = out[:, :, :h, :w]
out_list.append((out.cpu(), win))
torch.cuda.empty_cache()
self.wins = None
return self.merge(out_list)
def evaluate(ckpt_path, config_path='base.hrnetw32', use_tta=False):
cfg = import_config(config_path)
model_state_dict = load_model_state_dict_from_ckpt(ckpt_path)
log_dir = os.path.dirname(ckpt_path)
test_dataloader = make_dataloader(cfg['data']['test'])
model = make_model(cfg['model'])
model.load_state_dict(model_state_dict)
model.cuda()
model.eval()
metric_op = er.metric.PixelMetric(7, logdir=log_dir, logger=logger)
vis_dir = os.path.join(log_dir, 'vis-{}'.format(os.path.basename(ckpt_path)))
palette = np.array(list(COLOR_MAP.values())).reshape(-1).tolist()
viz_op = er.viz.VisualizeSegmm(vis_dir, palette)
segm_helper = SegmSlidingWinInference()
with torch.no_grad():
for idx, (img, gt) in enumerate(test_dataloader):
h, w = img.shape[2:4]
logging.info('Progress - [{} / {}] size = ({}, {})'.format(idx + 1, len(test_dataloader), h, w))
seg_helper = segm_helper.patch((h, w), patch_size=(args.patch_size, args.patch_size), stride=args.stride,
transforms=None)
pred = seg_helper.forward(model, img, size_divisor=32)
y_true = gt['cls']
y_true = y_true.cpu()
pred = pred.argmax(dim=1).cpu()
valid_inds = y_true != -1
metric_op.forward(y_true[valid_inds], pred[valid_inds])
for clsmap, imname in zip(pred, gt['fname']):
viz_op(clsmap.cpu().numpy().astype(np.uint8), imname.replace('tif', 'png'))
metric_op.summary_all()
torch.cuda.empty_cache()
if __name__ == '__main__':
# ckpt_path = './log/deeplabv3p.pth'
# config_path = 'baseline_loveda.deeplabv3p'
parser = argparse.ArgumentParser(description='Eval methods')
parser.add_argument('--ckpt_path', type=str,
help='ckpt path', default='./log/deeplabv3p.pth')
parser.add_argument('--config_path', type=str,
help='config path', default='baseline_loveda.deeplabv3p')
parser.add_argument('--tta', type=bool,
help='use tta', default=False)
parser.add_argument('--patch_size', type=int,
help='patch_size', default=512)
parser.add_argument('--stride', type=int,
help='stride', default=256)
args = parser.parse_args()
evaluate(args.ckpt_path, args.config_path, args.tta)