def __getitem__(self, idx, transform=True):
item = self.dataset[idx] #jst.py call function __getitem__
img = cv2.imread(item['path'])
label = item['label']
if transform:
line_strings = self.lane_to_linestrings(
item['old_anno']['lanes']
) # get from origin, to prevent the lane shape be resized twice
line_strings = LineStringsOnImage(
line_strings,
shape=img.shape) # lane and image resolution 1280*720
img, line_strings = self.transform(
image=img,
line_strings=line_strings) # lane and image resolution 640*360
#print(img.shape, line_strings.shape)
line_strings.clip_out_of_image_()
new_anno = {
'path': item['path'],
'lanes': self.linestrings_to_lanes(line_strings)
}
new_anno['categories'] = item[
'categories'] #categories has been fed into 'label'
label = self.transform_annotation(new_anno,
img_wh=(self.img_w,
self.img_h))['label']
img = img / 255.
if self.normalize:
img = (img - IMAGENET_MEAN) / IMAGENET_STD
img = self.to_tensor(img.astype(np.float32))
return (img, label, idx)
def __getitem__(self, idx, transform=False):
item = self._annotations[idx]
img = cv2.imread(item['path'])
label = item['label']
if transform:
line_strings = self.lane_to_linestrings(item['old_anno']['lanes'])
line_strings = LineStringsOnImage(line_strings, shape=img.shape)
img, line_strings = self.transform(image=img,
line_strings=line_strings)
line_strings.clip_out_of_image_()
new_anno = {
'path': item['path'],
'lanes': self.linestrings_to_lanes(line_strings)
}
new_anno['categories'] = item['categories']
label = self._transform_annotation(new_anno,
img_wh=(self.img_w,
self.img_h))['label']
img = img / 255.
if self.normalize:
img = (img - IMAGENET_MEAN) / IMAGENET_STD
img = self.to_tensor(img.astype(np.float32))
return (img, label, idx)
def test_augment_line_strings__kernel_size_is_two__no_keep_size(self):
from imgaug.augmentables.lines import LineString, LineStringsOnImage
ls = [LineString([(1.5, 1.5), (5.5, 1.5), (5.5, 5.5)])]
lsoi = LineStringsOnImage(ls, shape=(6, 6, 3))
expected = LineStringsOnImage([
LineString([(1.5/2, 1.5/2), (5.5/2, 1.5/2), (5.5/2, 5.5/2)])
], shape=(3, 3, 3))
self._test_augment_cbaoi__kernel_size_is_two__no_keep_size(
lsoi, expected, "augment_line_strings")
def test_line_strings_alignment(self):
from imgaug.augmentables.lines import LineString, LineStringsOnImage
lss = [LineString([(10, 10), (30, 10), (30, 30)])]
lsoi = LineStringsOnImage(lss, shape=(40, 40, 1))
lsoi_empty = LineStringsOnImage([], shape=(40, 40, 1))
self._test_cbaoi_alignment(lsoi, lsoi_empty, [[(10 / 2, 10 / 2),
(30 / 2, 10 / 2),
(30 / 2, 30 / 2)]],
[[(10, 10), (30, 10),
(30, 30)]], "augment_line_strings")
def __getitem__(self, idx):
item = self.dataset[idx]
img_org = cv2.imread(item['path'])
line_strings_org = self.lane_to_linestrings(item['old_anno']['lanes'])
line_strings_org = LineStringsOnImage(line_strings_org,
shape=img_org.shape)
for i in range(30):
img, line_strings = self.transform(image=img_org.copy(),
line_strings=line_strings_org)
line_strings.clip_out_of_image_()
new_anno = {
'path': item['path'],
'lanes': self.linestrings_to_lanes(line_strings)
}
try:
label = self.transform_annotation(new_anno,
img_wh=(self.img_w,
self.img_h))['label']
break
except:
if (i + 1) == 30:
self.logger.critical(
'Transform annotation failed 30 times :(')
exit()
img = img / 255.
if self.normalize:
img = (img - IMAGENET_MEAN) / IMAGENET_STD
img = self.to_tensor(img.astype(np.float32))
return (img, label, idx)
def _lane_argue(*, image, lane_src):
lines_tuple = [[(float(pt['x']), float(pt['y'])) for pt in line_spec] for line_spec in lane_src['Lines']]
lss = [ia_LineString(line_tuple_spec) for line_tuple_spec in lines_tuple]
lsoi = LineStringsOnImage(lss, shape=image.shape)
color_shift = iaa.OneOf([
iaa.GaussianBlur(sigma=(0.5, 1.5)),
iaa.LinearContrast((1.5, 1.5), per_channel=False),
iaa.Multiply((0.8, 1.2), per_channel=0.2),
iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.1 * 255), per_channel=0.5),
iaa.WithColorspace(to_colorspace=iaa.CSPACE_HSV, from_colorspace=iaa.CSPACE_RGB,
children=iaa.WithChannels(0, iaa.Multiply((0.7, 1.3)))),
iaa.WithColorspace(to_colorspace=iaa.CSPACE_HSV, from_colorspace=iaa.CSPACE_RGB,
children=iaa.WithChannels(1, iaa.Multiply((0.1, 2)))),
iaa.WithColorspace(to_colorspace=iaa.CSPACE_HSV, from_colorspace=iaa.CSPACE_RGB,
children=iaa.WithChannels(2, iaa.Multiply((0.5, 1.5)))),
])
posion_shift = iaa.SomeOf(4, [
iaa.Fliplr(),
iaa.Crop(percent=([0, 0.2], [0, 0.15], [0, 0], [0, 0.15]), keep_size=True),
iaa.TranslateX(px=(-16, 16)),
iaa.ShearX(shear=(-15, 15)),
iaa.Rotate(rotate=(-15, 15))
])
aug = iaa.Sequential([
iaa.Sometimes(p=0.6, then_list=color_shift),
iaa.Sometimes(p=0.6, then_list=posion_shift)
], random_order=True)
batch = ia.Batch(images=[image], line_strings=[lsoi])
batch_aug = list(aug.augment_batches([batch]))[0] # augment_batches returns a generator
image_aug = batch_aug.images_aug[0]
lsoi_aug = batch_aug.line_strings_aug[0]
lane_aug = [[dict(x=kpt.x, y=kpt.y) for kpt in shapely_line.to_keypoints()] for shapely_line in lsoi_aug]
return image_aug, dict(Lines=lane_aug)
def _test_augment_line_strings__kernel_size_differs(self, shape, shape_exp):
from imgaug.augmentables.lines import LineString, LineStringsOnImage
ls = [LineString([(1.5, 5.5), (5.5, 1.5), (5.5, 5.5)])]
lsoi = LineStringsOnImage(ls, shape=shape)
aug = self.augmenter(
(iap.Deterministic(3), iap.Deterministic(2)),
keep_size=False)
lsoi_aug = aug.augment_line_strings(lsoi)
expected = LineStringsOnImage(
[LineString([
((1.5/shape[1])*shape_exp[1], (5.5/shape[0])*shape_exp[0]),
((5.5/shape[1])*shape_exp[1], (1.5/shape[0])*shape_exp[0]),
((5.5/shape[1])*shape_exp[1], (5.5/shape[0])*shape_exp[0])
])],
shape=shape_exp)
assert_cbaois_equal(lsoi_aug, expected)
def dealFile(img_path, img_savedir, label_path, label_savedir):
basename = os.path.basename(img_path).split(".")[0]
img_savesubdir = img_savedir
label_savesubdir = label_savedir
if (not os.path.exists(img_savesubdir)):
os.makedirs(img_savesubdir)
if (not os.path.exists(label_savesubdir)):
os.makedirs(label_savesubdir)
save_imgname = basename + ".jpg"
save_labelname = basename + ".txt"
image = imageio.imread(img_path)
lines_vec = []
lines_tem = load_line(label_path)
lines_svec = []
for idx, line in enumerate(lines_tem):
line_pts = []
p1 = (line[1], line[2])
p2 = (line[3], line[4])
line_pts.append(p1)
line_pts.append(p2)
line_s = LineString(line_pts, label="{}".format(int(line[0])))
lines_svec.append(line_s)
if (len(lines_svec) < 25):
return
lsoi = LineStringsOnImage(lines_svec, shape=image.shape)
imageio.imwrite(os.path.join(img_savesubdir, save_imgname), image)
saveLineString(os.path.join(label_savesubdir, save_labelname), lsoi)
for t in range(10):
save_aug_imgname = save_imgname.split(".")[0] + "_" + str(t) + ".jpg"
save_aug_labelname = save_labelname.split(".")[0] + "_" + str(
t) + ".txt"
image_aug, lsoi_aug = seq(image=image, line_strings=lsoi)
lsoi_aug = lsoi_aug.remove_out_of_image()
lsoi_aug = lsoi_aug.clip_out_of_image()
line_len = len(lsoi_aug.line_strings)
if (line_len < 25):
continue
# ia.imshow(lsoi_aug.draw_on_image(image_aug, size=3))
imageio.imwrite(os.path.join(img_savesubdir, save_aug_imgname),
image_aug)
saveLineString(os.path.join(label_savesubdir, save_aug_labelname),
lsoi_aug)
img_len = len(os.listdir(img_savesubdir))
if (img_len < 2):
shutil.rmtree(img_savesubdir)
shutil.rmtree(label_savesubdir)
def test_empty_line_strings(self):
from imgaug.augmentables.lines import LineStringsOnImage
cbaoi = LineStringsOnImage([], shape=(5, 6, 3))
self._test_empty_cbaoi(cbaoi, "augment_line_strings")
def test_augment_line_strings__kernel_size_is_two__keep_size(self):
from imgaug.augmentables.lines import LineString, LineStringsOnImage
ls = [LineString([(0, 0), (2, 0), (2, 2)])]
lsoi = LineStringsOnImage(ls, shape=(6, 6, 3))
self._test_augment_cbaoi__kernel_size_is_two__keep_size(
lsoi, "augment_line_strings")
def _test_augment_line_strings__kernel_size_is_noop(self, kernel_size):
from imgaug.augmentables.lines import LineString, LineStringsOnImage
ls = [LineString([(1, 1), (2, 1), (2, 2)])]
lsoi = LineStringsOnImage(ls, shape=(6, 6, 3))
self._test_augment_cbaoi__kernel_size_is_noop(kernel_size, lsoi,
"augment_line_strings")
def kp_detection(db, k_ind):
data_rng = setup_configurations.data_rng
batch_size = setup_configurations.batch_size
input_size = db.configs["input_size"]
lighting = db.configs["lighting"]
rand_color = db.configs["rand_color"]
images = np.zeros((batch_size, 3, input_size[0], input_size[1]),
dtype=np.float32) # b, 3, H, W
masks = np.zeros((batch_size, 1, input_size[0], input_size[1]),
dtype=np.float32) # b, 1, H, W
gt_lanes = []
db_size = db.db_inds.size # 3268 | 2782
for b_ind in range(batch_size):
if k_ind == 0:
db.shuffle_inds()
db_ind = db.db_inds[k_ind]
k_ind = (k_ind + 1) % db_size
# reading ground truth
item = db.detections(db_ind) # all in the raw coordinate
img = cv2.imread(item['path'])
mask = np.ones((1, img.shape[0], img.shape[1], 1), dtype=np.bool)
label = item['label']
transform = True
if transform:
line_strings = db.lane_to_linestrings(item['old_anno']['lanes'])
line_strings = LineStringsOnImage(line_strings, shape=img.shape)
img, line_strings, mask = db.transform(image=img,
line_strings=line_strings,
segmentation_maps=mask)
line_strings.clip_out_of_image_()
new_anno = {
'path': item['path'],
'lanes': db.linestrings_to_lanes(line_strings)
}
new_anno['categories'] = item['categories']
label = db._transform_annotation(new_anno,
img_wh=(input_size[1],
input_size[0]))['label']
# clip polys
tgt_ids = label[:, 0]
label = label[tgt_ids > 0]
# make lower the same
label[:, 1][label[:, 1] < 0] = 1
label[:, 1][...] = np.min(label[:, 1])
label = np.stack([label] * batch_size, axis=0)
gt_lanes.append(torch.from_numpy(label.astype(np.float32)))
img = (img / 255.).astype(np.float32)
if rand_color:
color_jittering_(data_rng, img)
if lighting:
lighting_(data_rng, img, 0.1, db.eig_val, db.eig_vec)
normalize_(img, db.mean, db.std)
images[b_ind] = img.transpose((2, 0, 1))
masks[b_ind] = np.logical_not(mask[:, :, :, 0])
images = torch.from_numpy(images)
masks = torch.from_numpy(masks)
return {"xs": [images, masks], "ys": [images, *gt_lanes]}, k_ind
