def testFlip():
base_dir = Path.db_root_dir('rssrai')
image_dir = os.path.join(base_dir, 'test')
save_dir = save_path(os.path.join(base_dir, 'test_output'))
_img_path_list = glob(os.path.join(image_dir, '*.tif'))
img_name_list = [name.split('/')[-1] for name in _img_path_list]
pprint(img_name_list)
pprint(image_dir)
rssraiImage = RssraiTestOneImage(img_name_list[0], image_dir, save_dir, 10,
4)
type = "origin"
# type = "vertical"
# type = "horizontal"
print(rssraiImage.images[type].shape)
for dateSet in rssraiImage.get_slide_dataSet(rssraiImage.images[type]):
print(dateSet)
for i in dateSet:
shape = i['image'].shape
print(shape)
# i['image'] = torch.zeros(shape[0], rssraiImage.num_classes, shape[2], shape[3]).cuda()
# rssraiImage.fill_image(type, i)
rssraiImage.saveResultRGBImage()
def test_spilt_valid_image():
# import pandas as pd
# df = pd.read_csv("valid_set.csv")
# name_list = df["文件名"].values.tolist()
# print(name_list)
base_path = Path.db_root_dir("rssrai")
save_label_path = os.path.join(base_path, "split_valid_256", "label")
if not os.path.exists(save_label_path):
os.makedirs(save_label_path)
label_path = os.path.join(base_path, "split_valid", "label")
save_image_path = os.path.join(base_path, "split_valid_256", "img")
if not os.path.exists(save_image_path):
os.makedirs(save_image_path)
image_path = os.path.join(base_path, "split_valid", "img")
label_name_list = [
path_name.split("/")[-1]
for path_name in glob(os.path.join(label_path, "*"))
]
print(len(label_name_list))
for label_name in tqdm(label_name_list):
image_name = label_name.replace("_label", "")
# print(image_name)
# print(label_name)
split_image(label_path, label_name, save_label_path, mode="RGB")
split_image(image_path, image_name, save_image_path, mode="CMYK")
def test_one_merge_image():
base_path = Path.db_root_dir("rssrai")
# 图片
image_path = os.path.join(base_path, "split_test_256")
save_image_path = os.path.join(base_path, "merge_test", "img")
if not os.path.exists(save_image_path):
os.makedirs(save_image_path)
merge_image(image_path, "GF2_PMS1__20150902_L1A0001015646-MSS1.tif",
save_image_path, "CMYK")
def testOneImage():
base_path = Path.db_root_dir("rssrai")
path = '/home/arron/Documents/grey/Project_Rssrai/rssrai/train/img'
name = 'GF2_PMS1__20150212_L1A0000647768-MSS1.tif'
file_image = Image.open(os.path.join(path, name))
np_image = np.array(file_image)[:, :, 1:]
image = Image.fromarray(np_image.astype('uint8')).convert("RGB")
image.save(os.path.join(base_path, name))
def test_spilt_test_image():
base_path = Path.db_root_dir("rssrai")
# 图片
image_path = os.path.join(base_path, "test")
image_list = glob(os.path.join(image_path, "*.tif"))
save_image_path = os.path.join(base_path, "split_test_256", "img")
spilt_all_images(image_list,
save_image_path,
mode="CMYK",
output_image_h_w=(256, 256))
def test_merge_images():
import pandas as pd
base_path = Path.db_root_dir("rssrai")
df = pd.read_csv("test_name_list.csv")
name_list = df['name'].tolist()
# 图片
image_path = os.path.join(base_path, "temp_test", "img")
save_image_path = os.path.join(base_path, "merge_test", "img")
if not os.path.exists(save_image_path):
os.makedirs(save_image_path)
for name in tqdm(name_list):
merge_image(image_path, name, save_image_path, "CMYK")
def test_one_spilt_test_image():
base_path = Path.db_root_dir("rssrai")
image_path = os.path.join(base_path, "test")
name = 'GF2_PMS1__20150902_L1A0001015646-MSS1.tif'
save_image_path = os.path.join(base_path, "split_test_256")
if not os.path.exists(save_image_path):
os.makedirs(save_image_path)
# 图片
split_image(image_path,
name,
save_image_path,
mode="CMYK",
output_image_h_w=(256, 256))
def testData():
plt.rcParams['savefig.dpi'] = 500 # 图片像素
plt.rcParams['figure.dpi'] = 500 # 分辨率
test_path = os.path.join(Path().db_root_dir("rssrai"), "测试输出")
if not os.path.exists(test_path):
os.makedirs(test_path)
rssrai = Rssrai(type="train")
for i in rssrai:
pprint(i["image"].shape)
pprint(i["label"].shape)
break
data_loader = DataLoader(rssrai, batch_size=4, shuffle=True, num_workers=4)
for ii, sample in enumerate(data_loader):
print(sample['image'].shape)
sample['image'] = sample['image'][:, 1:, :, :]
for jj in range(sample["image"].size()[0]):
img = sample['image'].numpy()
gt = sample['label'].numpy()
img_tmp = np.transpose(img[jj], axes=[1, 2, 0])
tmp = gt[jj]
segmap = decode_segmap(tmp)
img_tmp *= rssrai.std[1:]
img_tmp += rssrai.mean[1:]
img_tmp *= 255.0
img_tmp = img_tmp.astype(np.uint8)
plt.figure()
plt.title('display')
plt.subplot(121)
plt.imshow(img_tmp)
plt.subplot(122)
plt.imshow(segmap)
# with open( f"{test_path}/rssrai-{ii}-{jj}.txt", "w" ) as f:
# f.write( str( img_tmp ) )
# f.write( str( tmp ) )
# f.write( str( segmap ) )
plt.savefig(f"{test_path}/rssrai-{ii}-{jj}.jpg")
plt.close('all')
if ii == 3:
break
plt.show(block=True)
def __init__(self,
type,
base_size=(513, 513),
crop_size=(256, 256),
base_dir=Path.db_root_dir('pascal')):
"""
:param base_dir: path to VOC dataset directory
:param type: train/val
:param transform: transform to apply
"""
assert type in ['train', 'val']
super().__init__()
self._base_dir = base_dir
self._image_dir = os.path.join(self._base_dir, 'JPEGImages')
self._cat_dir = os.path.join(self._base_dir, 'SegmentationClass')
self.split = type
self.base_size = base_size
self.crop_size = crop_size
self.mean = (0.485, 0.456, 0.406)
self.std = (0.229, 0.224, 0.225)
self.im_ids = []
self.images = []
self.categories = []
_splits_dir = os.path.join(self._base_dir, 'ImageSets', 'Segmentation')
with open(os.path.join(os.path.join(_splits_dir, self.split + '.txt')),
"r") as f:
lines = f.read().splitlines()
for ii, line in enumerate(lines):
_image = os.path.join(self._image_dir, line + ".jpg")
_cat = os.path.join(self._cat_dir, line + ".png")
assert os.path.isfile(_image)
assert os.path.isfile(_cat)
self.im_ids.append(line)
self.images.append(_image)
self.categories.append(_cat)
assert (len(self.images) == len(self.categories))
# Display stats
print('Number of images in {}: {:d}'.format(type, len(self.images)))
def test_spilt_train_image():
base_path = Path.db_root_dir("rssrai")
# 图片
image_path = os.path.join(base_path, "train", "img")
image_list = glob(os.path.join(image_path, "*.tif"))
save_image_path = os.path.join(base_path, "split_train", "img")
spilt_all_images(image_list,
save_image_path,
mode="CMYK",
output_image_h_w=(680, 720))
# 标签
label_path = os.path.join(base_path, "train", "label")
label_list = glob(os.path.join(label_path, "*.tif"))
save_label_path = os.path.join(base_path, "split_train", "label")
spilt_all_images(label_list,
save_label_path,
mode="RGB",
output_image_h_w=(680, 720))
def testGetValid():
base_path = Path.db_root_dir("rssrai")
label_path = os.path.join(base_path, "split_train", "label")
img_path = os.path.join(base_path, "split_train", "img")
valid_label_path = os.path.join(base_path, "split_valid", "label")
valid_img_path = os.path.join(base_path, "split_valid", "img")
shutil.rmtree(valid_label_path)
os.makedirs(valid_label_path)
shutil.rmtree(valid_img_path)
os.makedirs(valid_img_path)
label_name_list = [
path_name.split("/")[-1]
for path_name in glob(os.path.join(label_path, "*"))
]
random.shuffle(label_name_list)
print(len(label_name_list))
valid_label_name_list = label_name_list[850:]
pprint(len(valid_label_name_list))
for label_name in valid_label_name_list:
img_name = label_name.replace("_label", "")
print(valid_label_path, label_name)
print(valid_img_path, img_name)
shutil.move(os.path.join(label_path, label_name),
os.path.join(valid_label_path, label_name))
shutil.move(os.path.join(img_path, img_name),
os.path.join(valid_img_path, img_name))
def testSplitLabel():
print(color_name_map)
name_list = ["文件名"]
for _, v in color_name_map.items():
name_list.append(v)
print(name_list)
all_statistic_list = []
base_path = Path.db_root_dir("rssrai")
image_path = os.path.join(base_path, "split", "label")
from glob import glob
image_list = glob(os.path.join(image_path, "*.tif"))
# 多进程
pool = Pool(16)
for image in tqdm(image_list):
list = image.split("/")
path = "/".join(list[:-1])
name = list[-1]
result = pool.apply_async(statistic_label, args=(path, name))
all_statistic_list.append(result.get())
df = pd.DataFrame(all_statistic_list, columns=name_list)
print(df)
df.to_csv(os.path.join(base_path, "split_label.csv"))
del rssraiImage
del self.model
gc.collect()
if __name__ == "__main__":
args = Options().parse()
args.dataset = 'rssrai'
args.model = 'FCN'
args.backbone = 'resnet50'
args.check_point_id = 1
args.batch_size = 500
print(args)
now = f"{time.localtime(time.time()).tm_year}-{time.localtime(time.time()).tm_mon}-{time.localtime(time.time()).tm_mday}-"
now += f"{time.localtime(time.time()).tm_hour}-{time.localtime(time.time()).tm_min}-{time.localtime(time.time()).tm_sec}"
tester = Tester()
base_dir = Path.db_root_dir('rssrai')
image_dir = os.path.join(base_dir, 'train', 'img')
save_dir = save_path(
os.path.join(base_dir, f'test_output_model={args.model}_time={now}'))
_img_path_list = glob(os.path.join(image_dir, '*.tif'))
img_name_list = [name.split('/')[-1] for name in _img_path_list]
pprint(img_name_list)
for index, name in enumerate(img_name_list):
print(f"{index}:{name}")
tester.test(name, image_dir, save_dir)
def __init__(self,
type='train',
base_size=(512, 512),
crop_size=(256, 256),
base_dir=Path.db_root_dir('rssrai')):
assert type in ['train', 'valid', 'test']
super().__init__()
self._base_dir = base_dir
self.type = type
self.in_c = 4
self.mean = mean
self.std = std
self.crop_size = crop_size
self.base_size = base_size
self.im_ids = []
self.images = []
self.categories = []
# 加载数据
if self.type == 'train':
# train_csv = os.path.join(self._base_dir, 'train_set.csv')
# self._label_name_list = pd.read_csv(train_csv)["文件名"].values.tolist()
self._label_path_list = glob(
os.path.join(self._base_dir, 'split_train', 'label', '*.tif'))
# print(self._label_path_list)
self._label_name_list = [
name.split('/')[-1] for name in self._label_path_list
]
# print(self._label_name_list)
self._image_dir = os.path.join(self._base_dir, 'split_train',
'img')
self._label_dir = os.path.join(self._base_dir, 'split_train',
'label')
self.len = 20000
if self.type == 'valid':
self._label_path_list = glob(
os.path.join(self._base_dir, 'split_valid_256', 'label',
'*.tif'))
self._label_name_list = [
name.split('/')[-1] for name in self._label_path_list
]
self._image_dir = os.path.join(self._base_dir, 'split_valid_256',
'img')
self._label_dir = os.path.join(self._base_dir, 'split_valid_256',
'label')
# self._label_name_list = pd.read_csv( valid_csv )["文件名"].values.tolist()
self.len = len(self._label_name_list)
if self.type == 'test':
self._img_path_list = glob(
os.path.join(self._base_dir, 'split_test_256', 'img', '*.tif'))
self._img_name_list = [
name.split('/')[-1] for name in self._img_path_list
]
self._image_dir = os.path.join(self._base_dir, 'split_test_256',
'img')
self.len = len(self._img_path_list)