-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_floodnet.py
59 lines (47 loc) · 2.04 KB
/
train_floodnet.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
import ever as er
import torch
import numpy as np
import os
from data.floodnet import COLOR_MAP
from tqdm import tqdm
import random
from eval_window import SegmSlidingWinInference
er.registry.register_all()
def evaluate_cls_fn(self, test_dataloader, config=None):
self.model.eval()
classes = self.model.module.config.classes if self.model.module.config.classes != 1 else 2
metric_op = er.metric.PixelMetric(classes, logdir=self._model_dir, logger=self.logger)
vis_dir = os.path.join(self._model_dir, 'vis-{}'.format(self.checkpoint.global_step))
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 img, gt in tqdm(test_dataloader):
h, w = img.shape[2:4]
seg_helper = segm_helper.patch((h, w), patch_size=(1024, 1024), stride=512,
transforms=None)
pred = seg_helper.forward(self.model, img, size_divisor=32)
y_true = gt['cls']
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('jpg', 'png'))
metric_op.summary_all()
torch.cuda.empty_cache()
def register_evaluate_fn(launcher):
launcher.override_evaluate(evaluate_cls_fn)
def seed_torch(seed=2333):
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed) # if you are using multi-GPU.
torch.backends.cudnn.benchmark = True
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.enabled = True
if __name__ == '__main__':
seed_torch(2333)
trainer = er.trainer.get_trainer('th_amp_ddp')()
trainer.run(after_construct_launcher_callbacks=[register_evaluate_fn])