def __init__(self, semantic_version_dataroot: str,
instance_version_dataroot: str):
"""
Args:
- semantic_version_dataroot: from ADEChallengeData2016
- instance_verson_dataroot: from ADE20K_2016_07_26
"""
self.img_dir = f'{semantic_version_dataroot}/images'
self.ade20k_instance_dataroot = instance_version_dataroot
# following two maps are only used for relabeling original data
self.id_to_classname_map = get_dataloader_id_to_classname_map(
dataset_name='ade20k-151')
self.classname_to_id_map = get_classname_to_dataloaderid_map(
dataset_name='ade20k-151')
self.ade20k_split_nickname_dict = {
'train': 'training',
'val': 'validation'
}
# make it easy to look up an RGB file path, given just the filename stem
self.fname_to_rgbfpath_dict = {}
for split in ['train', 'val']:
ade20k_split_nickname = self.ade20k_split_nickname_dict[split]
rgb_fpaths = glob.glob(
f'{self.img_dir}/{ade20k_split_nickname}/*.jpg')
for rgb_fpath in rgb_fpaths:
fname_stem = Path(rgb_fpath).stem
assert fname_stem not in self.fname_to_rgbfpath_dict
self.fname_to_rgbfpath_dict[fname_stem] = rgb_fpath
# make it easy to look up a SEG RGB file path, given just the filename stem
self.fname_to_segrgbfpath_dict = {}
seg_rgb_fpaths = glob.glob(
f'{self.ade20k_instance_dataroot}/images/**/*seg.png',
recursive=True)
for seg_rgb_fpath in seg_rgb_fpaths:
fname_stem = Path(seg_rgb_fpath).stem
assert '_seg' in fname_stem[-4:]
# remove suffix now
fname_stem = fname_stem.replace('_seg', '')
self.fname_to_segrgbfpath_dict[fname_stem] = seg_rgb_fpath
def test_save_pred_vs_label_7tuple_uniform_random_preds():
"""
"""
data_dir = f'{TEST_DATA_ROOT}/Camvid_test_data'
img_fpath = f'{data_dir}/images/0016E5_08159.png'
label_fpath = f'{data_dir}/preds/0016E5_08159.png'
img_rgb = imageio.imread(img_fpath)
label_img = imageio.imread(label_fpath)
img_h, img_w = label_img.shape
pred_img = np.random.randint(0, 459, (img_h, img_w)).astype(np.uint16)
id_to_class_name_map = get_dataloader_id_to_classname_map(
'pascal-context-460')
save_fpath = f'{TEST_DATA_ROOT}/rand_549_temp.png'
save_pred_vs_label_7tuple(img_rgb, pred_img, label_img,
id_to_class_name_map, save_fpath)
os.remove(f'{TEST_DATA_ROOT}/rand_549_temp_pred_labels_palette.png')
def test_save_pred_vs_label_7tuple_shifted():
""" """
data_dir = f'{TEST_DATA_ROOT}/Camvid_test_data'
img_fpath = f'{data_dir}/images/0016E5_08159.png'
label_fpath = f'{data_dir}/preds/0016E5_08159.png'
img_rgb = imageio.imread(img_fpath)
label_img = imageio.imread(label_fpath)
pred_img = np.zeros_like(label_img)
pred_img[:, 150:] = label_img[:, :-150]
id_to_class_name_map = get_dataloader_id_to_classname_map(
'pascal-context-460')
save_fpath = f'{TEST_DATA_ROOT}/shifted_temp.png'
save_pred_vs_label_7tuple(img_rgb, pred_img, label_img,
id_to_class_name_map, save_fpath)
os.remove(f'{TEST_DATA_ROOT}/shifted_temp_pred_labels_palette.png')
def visual_sanitychecks():
"""
Save every 1000th image of each dataset, with classnames embedded.
"""
save_dir = 'temp_files/MSeg_verify'
for dname, d_info in infos.items():
print(f'Writing visual sanity checks for {dname}...')
if dname in [
'coco-panoptic-inst-201', 'mapillary-public66',
'ade20k-151-inst'
]:
continue # is in RGB format and not comparable
id_to_classname_map = get_dataloader_id_to_classname_map(dname)
splits = ['train', 'val']
split_lists = [d_info.trainlist, d_info.vallist]
for split, split_list in zip(splits, split_lists):
pairs = generate_all_img_label_pair_fpaths(d_info.dataroot,
split_list)
for i, (rgb_fpath, label_fpath) in enumerate(pairs[::1000]):
print(f'On {i*1000}/{len(pairs)} of {dname}')
rgb_img = cv2_imread_rgb(rgb_fpath)
label_img = imageio.imread(label_fpath)
fname_stem = Path(rgb_fpath).stem
save_fpath = f'{save_dir}/{dname}_{fname_stem}.jpg'
if rgb_img.ndim == 2:
# this image was grayscale
rgb_img = grayscale_to_color(rgb_img)
form_mask_triple_embedded_classnames(rgb_img,
label_img,
id_to_classname_map,
save_fpath,
save_to_disk=True)
def dump_pascalcontext_mat_files(pcontext_dst_dir: str) -> None:
"""
Convert PASCAL Context annotations from .mat files to .png
Args:
- pcontext_dst_dir: string represent absolute path to
PASCAL Context destination directory
Returns:
- None
"""
dataset_name = 'pascal-context-460'
id_to_class_name_map = get_dataloader_id_to_classname_map(
dataset_name, include_ignore_idx_cls=False)
save_dirname = 'Segmentation_GT_460cls'
png_save_dir = f'{pcontext_dst_dir}/{save_dirname}'
check_mkdir(png_save_dir)
# annotation files, stored as .mat files
mat_files_dir = f'{pcontext_dst_dir}/trainval'
mat_fpaths = glob.glob(f'{mat_files_dir}/*.mat')
for i, mat_fpath in enumerate(mat_fpaths):
if i % 500 == 0:
print(f'On {i}/{len(mat_fpaths)}')
fname_stem = Path(mat_fpath).stem
label_data = scipy.io.loadmat(mat_fpath)
label_img = label_data['LabelMap']
label_save_fpath = f'{png_save_dir}/{fname_stem}.png'
# Need uint16 to be able to exceed 256 value range
# there are up to 460 classes present.
label_img_uint16 = label_img.astype(np.uint16)
imageio.imwrite(label_save_fpath, label_img_uint16)
loaded_label_img = imageio.imread(label_save_fpath)
assert np.allclose(loaded_label_img, label_img)
def test_save_pred_vs_label_7tuple_100_strided_preds():
"""
"""
data_dir = f'{TEST_DATA_ROOT}/Camvid_test_data'
img_fpath = f'{data_dir}/images/0016E5_08159.png'
label_fpath = f'{data_dir}/preds/0016E5_08159.png'
img_rgb = imageio.imread(img_fpath)
label_img = imageio.imread(label_fpath)
img_h, img_w = label_img.shape
class_ids_to_sample = np.array([0, 100, 200, 300])
pred_img = np.random.choice(a=class_ids_to_sample,
size=(img_h, img_w)).astype(np.uint16)
id_to_class_name_map = get_dataloader_id_to_classname_map(
'pascal-context-460')
save_fpath = f'{TEST_DATA_ROOT}/100_strided_temp.png'
save_pred_vs_label_7tuple(img_rgb, pred_img, label_img,
id_to_class_name_map, save_fpath)
os.remove(f'{TEST_DATA_ROOT}/100_strided_temp_pred_labels_palette.png')
def __init__(self, dataroot: str, dname: str) -> None:
"""
"""
self.dataroot = dataroot
self.dname = dname
self.id_to_classname_map = get_dataloader_id_to_classname_map(dname)
def main():
"""
"""
save_dir = 'temp_files'
# beforehand
# orig_dname = 'cityscapes-19'
# orig_dataroot = '/export/share/Datasets/cityscapes'
# rel_dname = 'cityscapes-19'
# rel_dataroot = '/export/share/Datasets/MSeg/mseg_dataset/Cityscapes'
orig_dname = 'coco-panoptic-133'
orig_dname_lists = 'coco-panoptic-133-temp'
orig_dataroot = '/export/share/Datasets/COCO2017'
rel_dname = 'coco-panoptic-133'
rel_dataroot = '/export/share/Datasets/MSeg/mseg_dataset/COCOPanoptic'
orig_id_to_classname_map = get_dataloader_id_to_classname_map(orig_dname)
rel_id_to_classname_map = get_dataloader_id_to_classname_map(rel_dname)
for split in ['train', 'val']:
orig_split_txt_fpath = f'../mseg/dataset_lists/{orig_dname_lists}/list/{split}.txt'
orig_pairs = generate_all_img_label_pair_fpaths(orig_dataroot, orig_split_txt_fpath)
rel_split_txt_fpath = f'../mseg/dataset_lists/{rel_dname}/list/{split}.txt'
rel_pairs = generate_all_img_label_pair_fpaths(rel_dataroot, rel_split_txt_fpath)
for i in range(len(orig_pairs))[::100]:
orig_pair = orig_pairs[i]
orig_rgb_fpath, orig_label_fpath = orig_pair
orig_rgb_img = imageio.imread(orig_rgb_fpath)
orig_label_img = imageio.imread(orig_label_fpath)
rel_pair = rel_pairs[i]
rel_rgb_fpath, rel_label_fpath = rel_pair
rel_rgb_img = imageio.imread(rel_rgb_fpath)
rel_label_img = imageio.imread(rel_label_fpath)
if not np.allclose(orig_label_img,rel_label_img):
pdb.set_trace()
fname_stem = Path(orig_rgb_fpath).stem
orig_save_fpath = f'{save_dir}/{split}_{i}_noguide_orig_{fname_stem}.png'
orig_guide_save_fpath = f'{save_dir}/{split}_{i}_guide_orig_{fname_stem}.png'
rel_save_fpath = f'{save_dir}/{split}_{i}_noguide_rel_{fname_stem}.png'
rel_guide_save_fpath = f'{save_dir}/{split}_{i}_guide_rel_{fname_stem}.png'
if orig_rgb_img.ndim == 2:
# this image was grayscale
orig_rgb_img = grayscale_to_color(orig_rgb_img)
if rel_rgb_img.ndim == 2:
# this image was grayscale
rel_rgb_img = grayscale_to_color(rel_rgb_img)
#save_classnames_in_image(img_rgb, label_img, id_to_class_name_map, save_to_disk=True, save_fpath=save_fpath)
form_mask_triple_embedded_classnames(orig_rgb_img, orig_label_img, orig_id_to_classname_map, orig_save_fpath, save_to_disk=True)
save_mask_triple_with_color_guide(orig_rgb_img, orig_label_img, orig_id_to_classname_map,fname_stem,save_dir, orig_guide_save_fpath)
form_mask_triple_embedded_classnames(rel_rgb_img, rel_label_img, rel_id_to_classname_map, rel_save_fpath, save_to_disk=True)
save_mask_triple_with_color_guide(rel_rgb_img, rel_label_img, rel_id_to_classname_map,fname_stem,save_dir, rel_guide_save_fpath)
from tqdm import tqdm
from mseg.utils.names_utils import (get_classname_to_dataloaderid_map,
get_dataloader_id_to_classname_map)
"""
AutoNUE labels and classes to store, read, and write annotations.
In their hierarchy, we use 'id-type'='id' (the most fine-grained).
Uses PIllow to raster json polygons.
Ref:
https://github.com/AutoNUE/public-code/blob/master/preperation/createLabels.py
"""
name2labelid = get_classname_to_dataloaderid_map('idd-40',
include_ignore_idx_cls=False)
id2label = get_dataloader_id_to_classname_map('idd-40',
include_ignore_idx_cls=False)
# A point in a polygon
Point = namedtuple('Point', ['x', 'y'])
# Class that contains the information of a single annotated object
class CsObject:
# Constructor
def __init__(self):
# the label
self.label = ""
# the polygon as list of points
self.polygon = []
# the object ID
self.id = -1
