def __airbus_ship_parse__(self, label_file, image_file):
"""
(xmin, ymin, xmax, ymax)
"""
image_file_name = os.path.splitext(os.path.basename(image_file))[0]
save_image = wwtool.generate_image(800, 800, color=(0, 0, 0))
img = cv2.imread(image_file)
img_height, img_width, _ = img.shape
image_objects = airbus_ship_parse.airbus_ship_parse(os.path.basename(image_file))
objects = []
objects_small_save = []
total_object_num = len(image_objects)
small_object_num = 0
large_object_num = 0
total_object_num = 0
for image_object in image_objects:
object_struct = {}
object_struct_small = {}
xmin, ymin, xmax, ymax = image_object['bbox']
bbox_w = xmax - xmin
bbox_h = ymax - ymin
if bbox_w * bbox_h <= self.small_object_area:
continue
total_object_num += 1
if bbox_h * bbox_w <= small_size:
small_object_num += 1
if bbox_h * bbox_w >= large_object_size:
large_object_num += 1
object_struct['bbox'] = [xmin, ymin, bbox_w, bbox_h]
object_struct['segmentation'] = wwtool.bbox2pointobb([xmin, ymin, xmax, ymax])
object_struct['label'] = 1
object_struct_small = object_struct.copy()
object_struct_small['bbox'] = [xmin, ymin, xmax, ymax]
object_struct_small['label'] = 'ship'
objects_small_save.append(object_struct_small)
objects.append(object_struct)
if total_object_num > self.max_object_num_per_image:
self.max_object_num_per_image = total_object_num
if just_keep_small or generate_small_dataset:
if small_object_num >= total_object_num * small_object_rate and large_object_num < 1 and len(objects_small_save) > 0:
# print(os.path.join(dst_image_path, image_file_name + '.png'))
save_image[0:img_height, 0:img_width, :] = img
cv2.imwrite(os.path.join(dst_image_path, image_file_name + '.png'), save_image)
anno_file = os.path.join(dst_label_path, image_file_name + '.txt')
wwtool.simpletxt_dump(objects_small_save, anno_file)
return objects
else:
return []
else:
return objects
def split_image(self, image_fn):
if not image_fn.endswith('.jpg'):
return
image_file = os.path.join(self.image_path, image_fn)
shp_file = os.path.join(self.merged_shp_path, image_fn.replace('jpg', 'shp'))
geo_file = os.path.join(self.geo_path, image_fn.replace('jpg', 'png'))
file_name = os.path.splitext(os.path.basename(image_file))[0]
if not os.path.exists(shp_file):
return
img = cv2.imread(image_file)
geo_info = rio.open(geo_file)
objects = self.shp_parser(shp_file, geo_info)
masks = np.array([obj['segmentation'] for obj in objects])
subimages = wwtool.split_image(img, subsize=self.subimage_size, gap=self.gap)
subimage_coordinates = list(subimages.keys())
if masks.shape[0] == 0:
return
mask_centroids = []
for obj in objects:
geometry = obj['converted_polygon'].centroid
geo = geojson.Feature(geometry=geometry, properties={})
coordinate = geo.geometry["coordinates"]
coordinate[0], coordinate[1] = abs(coordinate[0]), abs(coordinate[1])
mask_centroids.append(coordinate)
mask_centroids = np.array(mask_centroids)
mask_centroids_ = mask_centroids.copy()
for subimage_coordinate in subimage_coordinates:
objects = []
mask_centroids_[:, 0] = mask_centroids[:, 0] - subimage_coordinate[0]
mask_centroids_[:, 1] = mask_centroids[:, 1] - subimage_coordinate[1]
cx_bool = np.logical_and(mask_centroids_[:, 0] >= 0, mask_centroids_[:, 0] < subimage_size)
cy_bool = np.logical_and(mask_centroids_[:, 1] >= 0, mask_centroids_[:, 1] < subimage_size)
subimage_masks = masks[np.logical_and(cx_bool, cy_bool)]
subimage_masks_ = []
for subimage_mask in subimage_masks:
subimage_mask_np = np.array(subimage_mask)
subimage_mask_np[0::2] = subimage_mask_np[0::2] - subimage_coordinate[0]
subimage_mask_np[1::2] = subimage_mask_np[1::2] - subimage_coordinate[1]
subimage_masks_.append(subimage_mask_np.tolist())
subimage_masks = subimage_masks_
if len(subimage_masks) == 0:
continue
img = subimages[subimage_coordinate]
if np.mean(img) == 0:
continue
label_save_file = os.path.join(self.label_save_path, '{}__{}_{}.txt'.format(file_name, subimage_coordinate[0], subimage_coordinate[1]))
image_save_file = os.path.join(self.image_save_path, '{}__{}_{}.png'.format(file_name, subimage_coordinate[0], subimage_coordinate[1]))
cv2.imwrite(image_save_file, img)
for subimage_mask in subimage_masks:
subimage_objects = dict()
subimage_objects['mask'] = subimage_mask
subimage_objects['label'] = 'building'
objects.append(subimage_objects)
wwtool.simpletxt_dump(objects, label_save_file, encode='mask')
bboxes_[:, 0] = bboxes[:, 0] - subimage_coordinate[0]
bboxes_[:, 1] = bboxes[:, 1] - subimage_coordinate[1]
cx_bool = np.logical_and(bboxes_[:, 0] >= 0,
bboxes_[:, 0] < 1024)
cy_bool = np.logical_and(bboxes_[:, 1] >= 0,
bboxes_[:, 1] < 1024)
subimage_bboxes = bboxes_[np.logical_and(cx_bool, cy_bool)]
if len(subimage_bboxes) == 0:
continue
img = subimages[subimage_coordinate]
if np.mean(img) == 0:
continue
label_save_file = os.path.join(
label_save_path,
'{}__{}_{}.txt'.format(file_name, subimage_coordinate[0],
subimage_coordinate[1]))
image_save_file = os.path.join(
image_save_path,
'{}__{}_{}.png'.format(file_name, subimage_coordinate[0],
subimage_coordinate[1]))
cv2.imwrite(image_save_file, img)
for subimage_bbox in subimage_bboxes:
subimage_objects = dict()
subimage_objects['bbox'] = subimage_bbox.tolist()
subimage_objects['label'] = 'ship'
objects.append(subimage_objects)
wwtool.simpletxt_dump(objects, label_save_file)
def split_image(self, image_fn):
if not image_fn.endswith('.jpg'):
return
image_file = os.path.join(self.image_path, image_fn)
shp_file = os.path.join(self.roof_shp_path,
image_fn.replace('jpg', 'shp'))
geo_file = image_file
file_name = os.path.splitext(os.path.basename(image_file))[0]
if not os.path.exists(shp_file):
return
img = cv2.imread(image_file)
geo_info = rio.open(geo_file)
objects = self.shp_parser(shp_file,
geo_info,
coord='pixel',
ignore_file=None,
merge_flag=True,
connection_mode='floor')
masks = np.array([obj['segmentation'] for obj in objects])
subimages = wwtool.split_image(img,
subsize=self.subimage_size,
gap=self.gap)
subimage_coordinates = list(subimages.keys())
if masks.shape[0] == 0:
return
mask_centroids = []
for obj in objects:
geometry = obj['converted_polygon'].centroid
geo = geojson.Feature(geometry=geometry, properties={})
coordinate = geo.geometry["coordinates"]
coordinate[0], coordinate[1] = abs(coordinate[0]), abs(
coordinate[1])
mask_centroids.append(coordinate)
mask_centroids = np.array(mask_centroids)
mask_centroids_ = mask_centroids.copy()
for subimage_coordinate in subimage_coordinates:
objects = []
mask_centroids_[:,
0] = mask_centroids[:, 0] - subimage_coordinate[0]
mask_centroids_[:,
1] = mask_centroids[:, 1] - subimage_coordinate[1]
cx_bool = np.logical_and(mask_centroids_[:, 0] >= 0,
mask_centroids_[:, 0] < subimage_size)
cy_bool = np.logical_and(mask_centroids_[:, 1] >= 0,
mask_centroids_[:, 1] < subimage_size)
subimage_masks = masks[np.logical_and(cx_bool, cy_bool)]
subimage_masks_ = []
for subimage_mask in subimage_masks:
if wwtool.mask2polygon(subimage_mask).area < 5:
continue
subimage_mask_np = np.array(subimage_mask)
subimage_mask_np[
0::2] = subimage_mask_np[0::2] - subimage_coordinate[0]
subimage_mask_np[
1::2] = subimage_mask_np[1::2] - subimage_coordinate[1]
subimage_masks_.append(subimage_mask_np.tolist())
subimage_masks = subimage_masks_
# cut the polygons by image boundary
subimage_masks = wwtool.clip_mask(subimage_masks,
image_size=(1024, 1024))
# judge whether to drop this subimage
drop_flag = self.drop_subimage(subimages,
subimage_coordinate,
subimage_masks,
show=self.show)
if drop_flag:
continue
img = subimages[subimage_coordinate]
label_save_file = os.path.join(
self.label_save_path,
'{}_{}__{}_{}.txt'.format(self.sub_imageset_fold, file_name,
subimage_coordinate[0],
subimage_coordinate[1]))
image_save_file = os.path.join(
self.image_save_path,
'{}_{}__{}_{}.png'.format(self.sub_imageset_fold, file_name,
subimage_coordinate[0],
subimage_coordinate[1]))
cv2.imwrite(image_save_file, img)
for subimage_mask in subimage_masks:
subimage_objects = dict()
subimage_objects['mask'] = subimage_mask
subimage_objects['label'] = 'building'
objects.append(subimage_objects)
wwtool.simpletxt_dump(objects, label_save_file, encode='mask')
origin_label_path = './data/dota-v1.5/v1/trainval/labelTxt-v1.5'
filtered_image_path = './data/dota-v1.5/v1/trainval_iltered/images'
filtered_label_path = './data/dota-v1.5/v1/trainval_filtered/labels'
wwtool.mkdir_or_exist(filtered_image_path)
wwtool.mkdir_or_exist(filtered_label_path)
filter_count = 1
progress_bar = mmcv.ProgressBar(len(os.listdir(origin_label_path)))
for label_name in os.listdir(origin_label_path):
image_objects = wwtool.dota_parse(os.path.join(origin_label_path, label_name))
filtered_objects = []
for image_object in image_objects:
if convert_classes[image_object['label']] == None:
filter_count += 1
continue
else:
image_object['label'] = convert_classes[image_object['label']]
filtered_objects.append(image_object)
if len(filtered_objects) > 0:
# img = cv2.imread(os.path.join(origin_image_path, os.path.splitext(label_name)[0] + image_format))
# save_image_file = os.path.join(filtered_image_path, os.path.splitext(label_name)[0] + '.png')
# print("Save image file: ", save_image_file)
# cv2.imwrite(save_image_file, img)
wwtool.simpletxt_dump(filtered_objects, os.path.join(filtered_label_path, os.path.splitext(label_name)[0] + '.txt'))
progress_bar.update()
print("Filter object counter: {}".format(filter_count))
subimage_mask_np[
0::2] = subimage_mask_np[0::2] - subimage_coordinate[0]
subimage_mask_np[
1::2] = subimage_mask_np[1::2] - subimage_coordinate[1]
subimage_masks_.append(subimage_mask_np.tolist())
subimage_masks = subimage_masks_
if len(subimage_masks) == 0:
continue
img = subimages[subimage_coordinate]
if np.mean(img) == 0:
continue
label_save_file = os.path.join(
label_save_path,
'{}__{}_{}.txt'.format(file_name, subimage_coordinate[0],
subimage_coordinate[1]))
image_save_file = os.path.join(
image_save_path,
'{}__{}_{}.png'.format(file_name, subimage_coordinate[0],
subimage_coordinate[1]))
cv2.imwrite(image_save_file, img)
for subimage_mask in subimage_masks:
subimage_objects = dict()
subimage_objects['mask'] = subimage_mask
subimage_objects['label'] = 'building'
objects.append(subimage_objects)
wwtool.simpletxt_dump(objects, label_save_file, encode='mask')
def __sn6_parse__(self, label_file, image_file):
"""
(xmin, ymin, xmax, ymax)
"""
objects = []
if self.groundtruth:
image_file_name = os.path.splitext(os.path.basename(image_file))[0]
image_name_postfix = image_file_name + '.tif'
if imageset == 'val':
if image_name_postfix in valset_image_names:
pass
else:
return []
if imageset == 'train':
if image_name_postfix in valset_image_names:
return []
else:
pass
save_image = wwtool.generate_image(800, 800, color=(0, 0, 0))
img = cv2.imread(image_file)
img_height, img_width, _ = img.shape
image_name = os.path.basename(image_file).split(
'SN6_{}_AOI_11_Rotterdam_{}_'.format(
imageset_file_name, data_source))[1].split('.tif')[0]
masks = sn6_parse.sn6_parse(image_name)
objects_small_save = []
total_object_num = len(masks)
small_object_num = 0
large_object_num = 0
total_object_num = 0
for mask in masks:
object_struct = {}
object_struct_small = {}
xmin, ymin, xmax, ymax = wwtool.pointobb2bbox(
mask['segmentation'])
bbox_w = xmax - xmin
bbox_h = ymax - ymin
total_object_num += 1
if bbox_h * bbox_w <= small_size:
small_object_num += 1
if bbox_h * bbox_w >= large_object_size:
large_object_num += 1
object_struct['bbox'] = [xmin, ymin, bbox_w, bbox_h]
object_struct['segmentation'] = mask['segmentation']
object_struct['label'] = 1
object_struct_small = object_struct.copy()
object_struct_small['bbox'] = [xmin, ymin, xmax, ymax]
object_struct_small['label'] = 'building'
objects_small_save.append(object_struct_small)
objects.append(object_struct)
if total_object_num > self.max_object_num_per_image:
self.max_object_num_per_image = total_object_num
if just_keep_small or generate_small_dataset:
if small_object_num >= total_object_num * small_object_rate and large_object_num < 1 and len(
objects_small_save) > 0:
# print(os.path.join(dst_image_path, image_file_name + '.png'))
save_image[0:img_height, 0:img_width, :] = img
cv2.imwrite(
os.path.join(dst_image_path, image_file_name + '.png'),
save_image)
anno_file = os.path.join(dst_label_path,
image_file_name + '.txt')
wwtool.simpletxt_dump(objects_small_save, anno_file)
return objects
else:
return []
else:
return objects
else:
obj_struct = {}
obj_struct['segmentation'] = [0, 0, 0, 0, 0, 0, 0, 0]
obj_struct['bbox'] = [0, 0, 0, 0]
obj_struct['label'] = 0
objects.append(obj_struct)
return objects