def main():
files = polygonsjson_path_list
# quit if we did not find anything
if not files:
printError("j) Did not find any files~!~!~!")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_color.png")
# do the conversion
try:
json2labelImg(f, dst, "color")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def evaluatePanoptic(gt_json_file,
gt_folder,
pred_json_file,
pred_folder,
resultsFile,
iou_type="segm",
dilation_ratio=0.005):
start_time = time.time()
with open(gt_json_file, 'r') as f:
gt_json = json.load(f)
with open(pred_json_file, 'r') as f:
pred_json = json.load(f)
categories = {el['id']: el for el in gt_json['categories']}
print("Evaluation panoptic segmentation metrics:")
print("Ground truth:")
print("\tSegmentation folder: {}".format(gt_folder))
print("\tJSON file: {}".format(gt_json_file))
print("Prediction:")
print("\tSegmentation folder: {}".format(pred_folder))
print("\tJSON file: {}".format(pred_json_file))
if not os.path.isdir(gt_folder):
printError(
"Folder {} with ground truth segmentations doesn't exist".format(
gt_folder))
if not os.path.isdir(pred_folder):
printError(
"Folder {} with predicted segmentations doesn't exist".format(
pred_folder))
pred_annotations = {el['image_id']: el for el in pred_json['annotations']}
matched_annotations_list = []
for gt_ann in gt_json['annotations']:
image_id = gt_ann['image_id']
if image_id not in pred_annotations:
raise Exception(
'no prediction for the image with id: {}'.format(image_id))
matched_annotations_list.append((gt_ann, pred_annotations[image_id]))
pq_stat = pq_compute_multi_core(matched_annotations_list, gt_folder,
pred_folder, categories, iou_type,
dilation_ratio)
results = average_pq(pq_stat, categories)
if resultsFile is not None:
with open(resultsFile, 'w') as f:
print("Saving computed results in {}".format(resultsFile))
json.dump(results, f, sort_keys=True, indent=4)
print_results(results, categories)
t_delta = time.time() - start_time
print("Time elapsed: {:0.2f} seconds".format(t_delta))
return results
def main():
# Where to look for Cityscapes
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
# MODIFIED: customized data path
cityscapesPath = os.path.join(
'/home/kywongaz/All_Models/pytorch-deeplab-xception/data/cityscapes/gtFine_trainvaltest'
)
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*",
"*_gt*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
# Where to look for Cityscapes
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
# how to search for all ground truth
cityscapesPath = '/home/pengsida/Datasets/cityscape'
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*",
"*_gt*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
f = '/home/pengsida/Datasets/cityscape/gtFine/val/frankfurt/frankfurt_000000_000294_gtFine_polygons.json'
# create the output filename
dst = f.replace("_polygons.json", "_instanceTrainIds.png")
# do the conversion
try:
json2instanceImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
# Where to look for Cityscapes
if 'DATASET_PATH' in os.environ:
datasetPath = os.environ['DATASET_PATH']
else:
datasetPath = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'..', '..')
# how to search for all ground truth
searchFine = os.path.join(datasetPath, "annotations", "*",
"*_polygons.json")
searchCoarse = os.path.join(datasetPath, "original_images", "*",
"*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
# Where to look for Cityscapes
if platform.system() == 'Windows':
cityscapesPath = 'D:/data/cityscapes'
else:
cityscapesPath = '/home/clova/data/cityscapes'
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*",
"*_gt*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
cityscapesPath = "/home/SENSETIME/parkchanho/Desktop/etc-repo"
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
#searchCoarse = os.path.join( cityscapesPath , "gtCoarse" , "*" , "*" , "*_gt*_polygons.json" )
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
#filesCoarse = glob.glob( searchCoarse )
#filesCoarse.sort()
# concatenate fine and coarse
files = filesFine
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
# Where to look for Cityscapes
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(os.path.dirname(os.path.realpath(__file__)),'..','..')
# how to search for all ground truth
searchFine = os.path.join( cityscapesPath , "gtFine" , "*" , "*" , "*_gt*_polygons.json" )
searchCoarse = os.path.join( cityscapesPath , "gtCoarse" , "*" , "*" , "*_gt*_polygons.json" )
# search files
filesFine = glob.glob( searchFine )
filesFine.sort()
filesCoarse = glob.glob( searchCoarse )
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError( "Did not find any files. Please consult the README." )
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format( progress * 100 / len(files) ), end=' ')
for f in files:
# create the output filename
dst = f.replace( "_polygons.json" , "_labelTrainIds.png" )
# do the conversion
try:
json2labelImg( f , dst , "trainIds" )
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format( progress * 100 / len(files) ), end=' ')
sys.stdout.flush()
def createTrainIdInstanceImgs(cityscapesPath, fine_fold_name,
coarse_fold_name):
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, fine_fold_name, "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, coarse_fold_name, "*", "*",
"*_gt*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_instanceTrainIds.png")
# do the conversion
try:
json2instanceImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def evaluatePanoptic(gt_json_file, gt_folder, pred_json_file, pred_folder, resultsFile):
start_time = time.time()
with open(gt_json_file, 'r') as f:
gt_json = json.load(f)
with open(pred_json_file, 'r') as f:
pred_json = json.load(f)
categories = {el['id']: el for el in gt_json['categories']}
print("Evaluation panoptic segmentation metrics:")
print("Ground truth:")
print("\tSegmentation folder: {}".format(gt_folder))
print("\tJSON file: {}".format(gt_json_file))
print("Prediction:")
print("\tSegmentation folder: {}".format(pred_folder))
print("\tJSON file: {}".format(pred_json_file))
if not os.path.isdir(gt_folder):
printError("Folder {} with ground truth segmentations doesn't exist".format(gt_folder))
if not os.path.isdir(pred_folder):
printError("Folder {} with predicted segmentations doesn't exist".format(pred_folder))
pred_annotations = {el['image_id']: el for el in pred_json['annotations']}
matched_annotations_list = []
for gt_ann in gt_json['annotations']:
image_id = gt_ann['image_id']
if image_id not in pred_annotations:
raise Exception('no prediction for the image with id: {}'.format(image_id))
matched_annotations_list.append((gt_ann, pred_annotations[image_id]))
pq_stat = pq_compute_multi_core(matched_annotations_list, gt_folder, pred_folder, categories)
results = average_pq(pq_stat, categories)
with open(resultsFile, 'w') as f:
print("Saving computed results in {}".format(resultsFile))
json.dump(results, f, sort_keys=True, indent=4)
print_results(results, categories)
t_delta = time.time() - start_time
print("Time elapsed: {:0.2f} seconds".format(t_delta))
return results
def convert2panoptic(cityscapesPath=None,
outputFolder=None,
useTrainId=False,
setNames=["val", "train", "test"]):
# i.21.3.18.1:35) Det2 에서는 지금 이 함수에서 useTrainId 값을 False 로 해준걸로 가정하고 처리해줌!!
# - Det2의 cityscapes_panoptic.py(cityscapes panoptic 데이터셋 레지스터해주는파일)의 _convert_category_id 함수는
# cityscapesscripts 의 createPanopticImgs.py (바로 지금 이 파일) 의 convert2panoptic 함수에서
# useTrainId=False 로 적용했을때(segment_info 의 "category_id" 가 카테고리의 trainId 가 아닌 그냥id로 셋팅됨)를 가정하고 작동하는거네.
# 그래서결국, segment_info["category_id"] 를 카테고리의 그냥id에서 trainId 로 바꿔주는거임.
# Where to look for Cityscapes
if cityscapesPath is None:
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
cityscapesPath = os.path.join(cityscapesPath, "gtFine")
if outputFolder is None:
outputFolder = cityscapesPath
categories = []
for label in labels:
if label.ignoreInEval:
continue
categories.append({
'id':
int(label.trainId) if useTrainId else int(label.id),
'name':
label.name,
'color':
label.color,
'supercategory':
label.category,
'isthing':
1 if label.hasInstances else 0
})
for setName in setNames:
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, setName, "*",
"*_instanceIds.png")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
files = filesFine
# quit if we did not find anything
if not files:
printError(
"Did not find any files for {} set using matching pattern {}. Please consult the README."
.format(setName, searchFine))
# a bit verbose
print("Converting {} annotation files for {} set.".format(
len(files), setName))
trainIfSuffix = "_trainId" if useTrainId else ""
outputBaseFile = "cityscapes_panoptic_{}{}".format(
setName, trainIfSuffix)
outFile = os.path.join(outputFolder, "{}.json".format(outputBaseFile))
print(
"Json file with the annotations in panoptic format will be saved in {}"
.format(outFile))
panopticFolder = os.path.join(outputFolder, outputBaseFile)
if not os.path.isdir(panopticFolder):
print("Creating folder {} for panoptic segmentation PNGs".format(
panopticFolder))
os.mkdir(panopticFolder)
print("Corresponding segmentations in .png format will be saved in {}".
format(panopticFolder))
images = []
annotations = []
for progress, f in enumerate(files):
originalFormat = np.array(Image.open(f))
fileName = os.path.basename(f)
imageId = fileName.replace("_gtFine_instanceIds.png", "")
inputFileName = fileName.replace("_instanceIds.png",
"_leftImg8bit.png")
outputFileName = fileName.replace("_instanceIds.png",
"_panoptic.png")
# image entry, id for image is its filename without extension
images.append({
"id": imageId,
"width": int(originalFormat.shape[1]),
"height": int(originalFormat.shape[0]),
"file_name": inputFileName
})
pan_format = np.zeros(
(originalFormat.shape[0], originalFormat.shape[1], 3),
dtype=np.uint8)
segmentIds = np.unique(originalFormat)
segmInfo = []
for segmentId in segmentIds:
if segmentId < 1000:
semanticId = segmentId
isCrowd = 1
else:
semanticId = segmentId // 1000
isCrowd = 0
labelInfo = id2label[semanticId]
categoryId = labelInfo.trainId if useTrainId else labelInfo.id
if labelInfo.ignoreInEval:
continue
if not labelInfo.hasInstances:
isCrowd = 0
mask = originalFormat == segmentId
color = [
segmentId % 256, segmentId // 256, segmentId // 256 // 256
]
pan_format[mask] = color
area = np.sum(mask) # segment area computation
# bbox computation for a segment
hor = np.sum(mask, axis=0)
hor_idx = np.nonzero(hor)[0]
x = hor_idx[0]
width = hor_idx[-1] - x + 1
vert = np.sum(mask, axis=1)
vert_idx = np.nonzero(vert)[0]
y = vert_idx[0]
height = vert_idx[-1] - y + 1
bbox = [int(x), int(y), int(width), int(height)]
segmInfo.append({
"id": int(segmentId),
"category_id": int(categoryId),
"area": int(area),
"bbox": bbox,
"iscrowd": isCrowd
})
annotations.append({
'image_id': imageId,
'file_name': outputFileName,
"segments_info": segmInfo
})
Image.fromarray(pan_format).save(
os.path.join(panopticFolder, outputFileName))
print("\rProgress: {:>3.2f} %".format(
(progress + 1) * 100 / len(files)),
end=' ')
sys.stdout.flush()
print("\nSaving the json file {}".format(outFile))
d = {
'images': images,
'annotations': annotations,
'categories': categories
}
with open(outFile, 'w') as f:
json.dump(d, f, sort_keys=True, indent=4)
def convert2panoptic(cityscapesPath=None,
outputFolder=None,
useTrainId=False,
setNames=["val", "train", "test"]):
# i.21.3.18.1:35) Det2 에서는 지금 이 함수에서 useTrainId 값을 False 로 해준걸로 가정하고 처리해줌!!
# - Det2의 cityscapes_panoptic.py(cityscapes panoptic 데이터셋 레지스터해주는파일)의 _convert_category_id 함수는
# cityscapesscripts 의 createPanopticImgs.py (바로 지금 이 파일에 해당하지) 의 convert2panoptic 함수에서
# useTrainId=False 로 적용했을때(segment_info 의 "category_id" 가 카테고리의 trainId 가 아닌 그냥id로 셋팅됨)를 가정하고 작동하는거네.
# 그래서결국, segment_info["category_id"] 를 카테고리의 그냥id에서 trainId 로 바꿔주는거임.
print(
f'j) <inputs shoud be> cityscapesPath:None, outputFolder:None, useTrainId:False, setNames:[\'train\', \'val\']'
)
print(
f'j) <actual inputs> cityscapesPath:{cityscapesPath}, outputFolder:{outputFolder}, useTrainId:{useTrainId}, setNames:{setNames}'
)
# Where to look for Cityscapes
if cityscapesPath is None:
cityscapesPath = CS_ROOTDIRPATH_J
if outputFolder is None:
outputFolder = os.path.join(CS_ROOTDIRPATH_J, "gt")
categories = [] # i. ######################################
for label in labels:
if label.ignoreInEval:
continue
categories.append({
'id':
int(label.trainId) if useTrainId else int(label.id),
'name':
label.name,
'color':
label.color,
'supercategory':
label.category,
'isthing':
1 if label.hasInstances else 0
})
# i. train, val 등의 폴더에 대해서.
for setName in setNames:
# how to search for all ground truth
forSearchInstanceIdsPngJ = os.path.join(cityscapesPath, "gt", setName,
"*_instanceIds.png")
# search files
instanceIdsPngPath_list = glob.glob(forSearchInstanceIdsPngJ)
instanceIdsPngPath_list.sort()
files = instanceIdsPngPath_list
# quit if we did not find anything
if not files:
printError(
"j) Did not find any files for {} set using matching pattern {} !!!"
.format(setName, forSearchInstanceIdsPngJ))
# a bit verbose
print(
"j) Converting {} annotation files(~~instanceIds.png) for {} set.".
format(len(files), setName))
trainIfSuffix = "_trainId" if useTrainId else ""
# outputBaseFile = "cityscapes_panoptic_{}{}".format(setName, trainIfSuffix)
outputBaseNameJ = "J_cocoformat_panoptic_{}{}".format(
setName, trainIfSuffix)
outAnnoJsonPathJ = os.path.join(outputFolder,
"{}.json".format(outputBaseNameJ))
print(
"Json file with the annotations in panoptic format will be saved in {}"
.format(outAnnoJsonPathJ))
panopticFolder = os.path.join(outputFolder, outputBaseNameJ)
if not os.path.isdir(panopticFolder):
print("Creating folder {} for panoptic segmentation PNGs".format(
panopticFolder))
os.mkdir(panopticFolder)
print("Corresponding segmentations in .png format will be saved in {}".
format(panopticFolder))
images = [] # i. ######################################
annotations = [] # i. ######################################
for progress, f in enumerate(files):
cs_annoPng_arrJ = np.array(
Image.open(f)) # i. f 는 path/to/~~instanceIds.png
print(f'j) cs_annoPng_arrJ.shape: {cs_annoPng_arrJ.shape}'
) # i. ex: (976, 1976)
fileName = os.path.basename(f)
# imageId = fileName.replace("_gtFine_instanceIds.png", "") # i. 변경필요 ###########
# inputFileName = fileName.replace("_instanceIds.png", "_leftImg8bit.png") # i. 변경필요 ###########
# outputFileName = fileName.replace("_instanceIds.png", "_panoptic.png")
# # fileName ex: imp2_0_instanceIds.png, imp4_120_instanceIds.png (imp{A}_{00B}_instanceIds.png 이런식. A:2,3,4, B:0~3자리수)
# # i.21.3.8.오후쯤) ->여기서 A00B 이런식으로 이미지id 만들어줄거임. B가 두자리면 A0bb, 세자리면 Abbb 이런식으로.
# # i.21.3.10.19:37) 굳이이렇게할필요없음. 이미지id 는 그냥 문자열 숫자열 섞여있어도 상관없음.
# # 그리고 Det2 의 데이터셋레지스터하는 코드 사용하려면.. 이미지id를 베이스네임?같은식으로 좀 맞춰줘야해서.. 암튼 이코드부분 수정필요.
# # ->바로아래에서수정함/21.3.16.11:42.
# implDatasetGroupNumJ = fileName[3] # "2", "4"
# implSubNumJ = fileName[len("impX_"):-len("_instanceIds.png")] # "0", "120"
# imageIdJ = implDatasetGroupNumJ + (3-len(implSubNumJ))*"0" + implSubNumJ # "2000", "4120" (A00B 이런식)
# inputImgFileNameJ = fileName.replace("_instanceIds.png", ".jpg")
# outAnnoPngNameJ = fileName.replace("_instanceIds.png", "_panopticAnno.png")
# i.21.3.16.11:37) 내가 기존엔 바로위코드처럼 이미지id 를 A00B이런식으로 만들어줬었는데,
# 굳이그럴필요없고 그냥 'imp2_0' 이런식으로 스트링으로 해줘도돼서, 그렇게하기로했음.
imageIdJ = fileName[:-len("_instanceIds.png")] # 'imp2_0' 이런식.
inputImgFileNameJ = fileName.replace("_instanceIds.png", ".jpg")
outAnnoPngNameJ = fileName.replace("_instanceIds.png",
"_panopticAnno.png")
# image entry, id for image is its filename without extension
images.append({
"id": imageIdJ,
"width": int(cs_annoPng_arrJ.shape[1]),
"height": int(cs_annoPng_arrJ.shape[0]),
"file_name": inputImgFileNameJ
})
# i.21.3.18.11:37) 여기서 coco_annoPng_arrJ 의 모든픽셀을 0으로 초기화해서 시작하는데,
# ignoreInEval 이 True 인 카테고리들은 그려지지않고 스킵되니까 초기화상태의 값인 [0,0,0] 이 그대로 백그라운드 픽셀의 값이 됨.
# 따라서, foreground 카테고리중에 id값을 0으로 해준게 있으면 256진법 RGB 로 변환시 [0,0,0] 이 돼버려서 백그라운드랑 똑같아져버리는 문제 발생.
coco_annoPng_arrJ = np.zeros(
(cs_annoPng_arrJ.shape[0], cs_annoPng_arrJ.shape[1], 3),
dtype=np.uint8 # i.21.3.9.8:04) Unsigned integer 0 to 255
)
print(f'j) coco_annoPng_arrJ.shape: {coco_annoPng_arrJ.shape}'
) # i. ex: (976, 1976, 3)
segmentIds = np.unique(
cs_annoPng_arrJ
) # i. ex: [ 0 1 2 3 4 5 6 7000 7001 7002 7003 7004 7005 7006 8000 8001 8002 8003 9000]
print(
f'j) segmentIds = np.unique(cs_annoPng_arrJ): {np.unique(cs_annoPng_arrJ)}'
)
segmsInfo = []
# i.21.3.8.00:25) 여기서 segmentIds 를 z-order에 맞게 정렬해줘야겠네.
# ->아니지. 지금 cs_annoPng_arrJ(~~instanceIds.png 를 읽어들인거)은 이미 내가정해준 zorder대로 그려진상태니까
# 이제는 zorder 상관할필요가 없겠네. 그냥 2차원 이미지일 뿐이니까.
for segmentId in segmentIds:
if segmentId < 1000: # i. id값이 1000미만이면, stuff(COCO형식에서 stuff는 iscrowd의미없고 기본적으로 0임)이거나, iscrowd=1인 thing임. 참고로 iscrowd 는 COCO형식에 나오는 값./21.3.9.9:38.
semanticId = segmentId
isCrowd = 1
else: # i. id값이 1000이상이면, iscrowd=0 인 thing임./21.3.9.9:38.
semanticId = segmentId // 1000
isCrowd = 0
# # i.21.3.17.23:06) 내가 'unlabeled_Label' 을 없애줬고 백그라운드의 segmentId 값은 내가정해준대로 255 일것이므로,
# # 바로아래의 labelInfo = id2label[semanticId] 가 실행되면 KeyError: 255 가 발생함.
# # 근데 어차피 기존 cityscapes 의 labels.py 대로라고 해도, label 의 ignoreInEval 이 True 일 경우 continue 해서 무시해주고있음(아래 보면 나오지).
# # 즉, 백그라운드는 걍 무시하면 됨. 따라서 semanticId(=segmentId) 값이 255이면 continue 해줌.
# # i.21.3.18.10:06) ->백그라운드는 무시하면 되는건 맞는데, mandible 의 id값을 0으로 해버려서 요아래 color 정해줄때
# # mandible 의 color 가 [0,0,0] 으로 돼버려서, mandible 이랑 백그라운드가 모두 [0,0,0] 으로 돼버림.
# # (위에서 coco_annoPng_arrJ 만들어줄때 모든원소들을 0으로 초기화해서 시작하니까 백그라운드 픽셀들은 기본적으로 [0,0,0] 인거지.)
# # 그래서 걍 다시 바꾸는중.
# if semanticId == 255:
# continue
labelInfo = id2label[semanticId]
# i. Det2 에서는 여기서 useTrainId 가 False 인걸로 가정하고 처리함.
# 즉, 여기서 categoryId 가 trainId 말고 그냥 id 인것으로 가정하고 처리함./21.3.18.9:57.
categoryId = labelInfo.trainId if useTrainId else labelInfo.id
if labelInfo.ignoreInEval: # i. <- 요거 기존코드인데, ignoreInEval 이 True 면 걍 continue 해서 무시하는것을 볼수있음./21.3.17.23:14.
continue
if not labelInfo.hasInstances: # i. stuff면, iscrowd 의미없고 기본적으로 0임.(COCO형식 참고하삼)/21.3.9.9:45.
isCrowd = 0
mask = cs_annoPng_arrJ == segmentId
# print(f'j) mask.shape: {mask.shape}') # i. ex: (976, 1976)
# color = [segmentId % 256, segmentId // 256, segmentId // 256 // 256]
# i.21.3.8.22:28)->요게 기존 코드. 잘못됐음. cityscapes 데이터셋은 클래스가 35갠가 뿐이라 이렇게해도 문제되진 않지만,
# 만약 클래스갯수가 65개고 인스턴스갯수가 엄청많다거나, 클래스갯수가 66개 이상이된다거나 하면 문제됨. 현 cityscapes 의 id정해주는방식이라면.
# 즉, 예를들어 id값이 65900(클래스번호65, 해당클래스의 901번째 인스턴스)라든가, id값이 66000(클래스번호66, 해당클래스의 0번째인스턴스)라든가 이럴경우 문제됨.
# (참고1: 256^2=65536)
# (참고2: COCO panoptic 형식에서 id=R+G*256+B*256^2, RGB는 어노png파일의 각 픽셀의 값.)
color = [
segmentId % 256, segmentId % (256 * 256) // 256,
segmentId % (256 * 256 * 256) // (256 * 256)
]
print(f'j) id:{segmentId} -> color(RGB):{color}')
# i.21.3.8.22:28)->요게 내가 수정한거. 세번쨋놈은 그냥 segmentId//(256*256)
# 또는 segmentId//256/256 으로 해도 되지만(id가 256^3보다 작을것이라서), 일반화를 위해 저렇게 적었음. 규칙성도 눈에잘보이고.
#
# i.21.3.10.19:44) cocodataset깃헙보면 cocoapi 말고도 panopticapi 라고 있음.
# 거기서 panopticapi.utils.id2rgb 함수가 바로 위의 변환이랑 내내 같은걸 해주고있음. (Det2 문서의 "pan_seg_file_name" 설명에 나오는 함수)
# 거깄는 id2rgb 함수는 2차원 id맵(numpy어레이)을 넣으면 RGB맵으로 변환해주고, 그냥 하나의 id값만 넣으면 [R,G,B] 리스트를 반환해줌.
coco_annoPng_arrJ[
mask] = color # i. ->coco_annoPng_arrJ 은 HxWx3, mask 는 HxW. 그래도 상관없지.
# ->color 는 list 지만, 일케해줘도 상관x. 넘파이어레이로 됨.
area = np.sum(mask) # segment area computation
# bbox computation for a segment
# i.21.3.20.21:42) 참고로, 지금 이 계산에서는, 만약 stuff 라서 원 인풋이미지상에서 되게 여기저기 넓게 분포한다해도, bbox는 그걸 다 커버할수있는 최소의 박스로 만들어짐.
# 그러니까 예를들어 원 인풋이미지에서 좌하 끝부분 코너, 우상 끝부분 코너 위치에 어던 stuff 가 있다치면, 예를들어 뭐 grass 가 있다면,
# 지금 여기서의 bbox는 그걸 다 커버하니까 엄청 커짐. 거의 뭐 이미지 전체를 뒤덮는 bbox 가 되겠지.
hor = np.sum(mask, axis=0)
hor_idx = np.nonzero(hor)[0]
x = hor_idx[0]
width = hor_idx[-1] - x + 1
vert = np.sum(mask, axis=1)
vert_idx = np.nonzero(vert)[0]
y = vert_idx[0]
height = vert_idx[-1] - y + 1
bbox = [int(x), int(y), int(width), int(height)]
# i.21.3.20.21:47) 참고로, thing 카테고리들과 달리, stuff 카테고리들의 경우
# 여기저기 산발적으로 여러개의 세그멘트들이 있다해도 그게 다 똑같은 id값을 가지니까,
# 그 모든 세그먼트들이 하나의 "segment_info"(지금 여기서 segmsInfo 에 append 해주는 dict) 에 대응되게 됨.
# 참고: createPanopticImgs.py 에선 원래 요 변수명 segmInfo 였는데,
# s 붙고 안붙고는 매우 중요하기때문에, s 붙여서 segmsInfo 로 내가 바꿔줬음.
segmsInfo.append({ # i. segmentId 는 ~~instanceIds.png 의 각 픽셀값.
"id": int(segmentId),
# i. Det2에선 이값이 trainId 아닌 그냥id 인걸로 가정하고,
# 이값을 trainId 로 바꿔서 Det2형식으로 만듦(데이터셋 레지스터해주는 cityscapse_panoptic.py 에서.) /21.3.18.10:00.
"category_id": int(categoryId),
"area": int(area),
"bbox": bbox,
"iscrowd": isCrowd
})
annotations.append({
'image_id': imageIdJ,
'file_name': outAnnoPngNameJ,
"segments_info": segmsInfo
})
Image.fromarray(coco_annoPng_arrJ).save(
os.path.join(panopticFolder, outAnnoPngNameJ))
print("\rProgress: {:>3.2f} %".format(
(progress + 1) * 100 / len(files)),
end=' ')
sys.stdout.flush()
print("\nSaving the json file {}".format(outAnnoJsonPathJ))
d = {
'images': images,
'annotations': annotations,
'categories': categories
}
with open(outAnnoJsonPathJ, 'w') as f:
json.dump(d, f, sort_keys=True, indent=4)
def main():
# Where to look for Cityscapes
if 'BraTS_DATASET' in os.environ:
bratsPath = os.environ['BraTS_DATASET']
else:
bratsPath = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'..', '..')
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*",
"*_gt*_polygons.json")
# searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*", "*_gt*_polygons.json")
# ot_path = glob.glob(os.path.join(os.path.join(bratsPath, "Training", "*", "*"), "*", "*OT*.mha"))
# t1_path = glob.glob(os.path.join(os.path.join(bratsPath, "Training", "*", "*"), "*", "*T1*.mha"))
# t1c_path = glob.glob(os.path.join(os.path.join(bratsPath, "Training", "*", "*"), "*", "*T1c*.mha"))
# flair_path = glob.glob(os.path.join(os.path.join(bratsPath, "Training", "*", "*"), "*", "*Flair*.mha"))
# t2_path = glob.glob(os.path.join(os.path.join(bratsPath, "Training", "*", "*"), "*", "*T2*.mha"))
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
# filesCoarse = glob.glob( searchCoarse )
# filesCoarse.sort()
# concatenate fine and coarse
files = filesFine
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def main():
cityscapesPath = os.environ.get(
'CITYSCAPES_DATASET',
os.path.join(os.path.dirname(os.path.realpath(__file__)), '..', '..'))
gtJsonFile = os.path.join(cityscapesPath, "gtFine",
"cityscapes_panoptic_val.json")
predictionPath = os.environ.get('CITYSCAPES_RESULTS',
os.path.join(cityscapesPath, "results"))
predictionJsonFile = os.path.join(predictionPath,
"cityscapes_panoptic_val.json")
parser = argparse.ArgumentParser()
parser.add_argument(
"--gt-json-file",
dest="gtJsonFile",
help=
'''path to json file that contains ground truth in COCO panoptic format.
By default it is $CITYSCAPES_DATASET/gtFine/cityscapes_panoptic_val.json.
''',
default=gtJsonFile,
type=str)
parser.add_argument(
"--gt-folder",
dest="gtFolder",
help='''path to folder that contains ground truth *.png files. If the
argument is not provided this script will look for the *.png files in
'name' if --gt-json-file set to 'name.json'.
''',
default=None,
type=str)
parser.add_argument(
"--prediction-json-file",
dest="predictionJsonFile",
help=
'''path to json file that contains prediction in COCO panoptic format.
By default is either $CITYSCAPES_RESULTS/cityscapes_panoptic_val.json
or $CITYSCAPES_DATASET/results/cityscapes_panoptic_val.json if
$CITYSCAPES_RESULTS is not set.
''',
default=predictionJsonFile,
type=str)
parser.add_argument(
"--prediction-folder",
dest="predictionFolder",
help='''path to folder that contains prediction *.png files. If the
argument is not provided this script will look for the *.png files in
'name' if --prediction-json-file set to 'name.json'.
''',
default=None,
type=str)
resultFile = "resultPanopticSemanticLabeling.json"
parser.add_argument(
"--results_file",
dest="resultsFile",
help="File to store computed panoptic quality. Default: {}".format(
resultFile),
default=resultFile,
type=str)
parser.add_argument("--iou-type", default="segm")
parser.add_argument("--dilation-ratio", default="0.005", type=float)
args = parser.parse_args()
if not os.path.isfile(args.gtJsonFile):
printError(
"Could not find a ground truth json file in {}. Please run the script with '--help'"
.format(args.gtJsonFile))
if args.gtFolder is None:
args.gtFolder = os.path.splitext(args.gtJsonFile)[0]
if not os.path.isfile(args.predictionJsonFile):
printError(
"Could not find a prediction json file in {}. Please run the script with '--help'"
.format(args.predictionJsonFile))
if args.predictionFolder is None:
args.predictionFolder = os.path.splitext(args.predictionJsonFile)[0]
evaluatePanoptic(args.gtJsonFile, args.gtFolder, args.predictionJsonFile,
args.predictionFolder, args.resultsFile, args.iou_type,
args.dilation_ratio)
return
def main():
# Where to look for Cityscapes
'''
__file__ 得到当前文件路径
但是若按绝对路径执行该文件，得到绝对路径
若按相对路径，或者在sys.path下执行，则得到相对路径。
为了保证得到绝对路径，用os.path.realpath()
os.path.dirname获得该文件所在的文件夹名称
os.path.join，组合成 目录/../..表示当前目录网上两次。
'''
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET'] #
else:
cityscapesPath = os.path.join(
os.path.dirname(os.path.realpath(__file__)), '..', '..')
# how to search for all ground truth
'''
在文件往上两个的目录，即cityscapes根目录下，找gtFine的目录。
加上*方便glob
'''
searchFine = os.path.join(cityscapesPath, "gtFine", "*", "*",
"*_gt*_polygons.json")
searchCoarse = os.path.join(cityscapesPath, "gtCoarse", "*", "*",
"*_gt*_polygons.json")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
filesCoarse = glob.glob(searchCoarse)
filesCoarse.sort()
# concatenate fine and coarse
files = filesFine + filesCoarse
# files = filesFine # use this line if fine is enough for now.
# quit if we did not find anything
'''
在python中 None, False, 空字符串"", 0, 空列表[], 空字典{}, 空元组()都相当于False
这里这个检查不太好，因为没检查图片为空的情形，只检查了label
'''
if not files:
printError("Did not find any files. Please consult the README.")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
'''
替换并建一个新的list放图片。这里是选其中labelTrainIds这个类型
'''
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status 更新进度条
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
# 输出刷新
sys.stdout.flush()
createPanopticImgs.convert2panoptic(pred_in_path, pred_out_path, True)
cityscapesPath = os.environ.get(
'CITYSCAPES_DATASET',
os.path.join(os.path.dirname(os.path.realpath(__file__)), '..', '..'))
gtJsonFile = os.path.join(cityscapesPath, "gtFine",
"cityscapes_panoptic_val.json")
predictionPath = os.environ.get('CITYSCAPES_RESULTS',
os.path.join(cityscapesPath, "results"))
predictionJsonFile = os.path.join(predictionPath,
"cityscapes_panoptic_val.json")
if not os.path.isfile(gt_jsonfile):
printError(
"Could not find a ground truth json file in {}. Please run the script with '--help'"
.format(gt_jsonfile))
if gt_folder is None:
gt_folder = os.path.splitext(gt_jsonfile)[0]
if not os.path.isfile(pred_jsonfile):
printError(
"Could not find a prediction json file in {}. Please run the script with '--help'"
.format(pred_jsonfile))
if pred_folder is None:
pred_folder = os.path.splitext(pred_jsonfile)[0]
evalPanopticSemanticLabeling.evaluatePanoptic(gt_jsonfile, gt_folder,
pred_jsonfile, pred_folder,
result)
def main():
cityscapesPath = os.environ.get(
'CITYSCAPES_DATASET', os.path.join(os.path.dirname(os.path.realpath(__file__)),'..','..')
)
gtJsonFile = os.path.join(cityscapesPath, "gtFine", "cityscapes_panoptic_val.json")
predictionPath = os.environ.get(
'CITYSCAPES_RESULTS',
os.path.join(cityscapesPath, "results")
)
predictionJsonFile = os.path.join(predictionPath, "cityscapes_panoptic_val.json")
parser = argparse.ArgumentParser()
parser.add_argument("--gt-json-file",
dest="gtJsonFile",
help= '''path to json file that contains ground truth in COCO panoptic format.
By default it is $CITYSCAPES_DATASET/gtFine/cityscapes_panoptic_val.json.
''',
default=gtJsonFile,
type=str)
parser.add_argument("--gt-folder",
dest="gtFolder",
help= '''path to folder that contains ground truth *.png files. If the
argument is not provided this script will look for the *.png files in
'name' if --gt-json-file set to 'name.json'.
''',
default=None,
type=str)
parser.add_argument("--prediction-json-file",
dest="predictionJsonFile",
help='''path to json file that contains prediction in COCO panoptic format.
By default is either $CITYSCAPES_RESULTS/cityscapes_panoptic_val.json
or $CITYSCAPES_DATASET/results/cityscapes_panoptic_val.json if
$CITYSCAPES_RESULTS is not set.
''',
default=predictionJsonFile,
type=str)
parser.add_argument("--prediction-folder",
dest="predictionFolder",
help='''path to folder that contains prediction *.png files. If the
argument is not provided this script will look for the *.png files in
'name' if --prediction-json-file set to 'name.json'.
''',
default=None,
type=str)
resultFile = "resultPanopticSemanticLabeling.json"
parser.add_argument("--results_file",
dest="resultsFile",
help="File to store computed panoptic quality. Default: {}".format(resultFile),
default=resultFile,
type=str)
args = parser.parse_args()
if not os.path.isfile(args.gtJsonFile):
printError("Could not find a ground truth json file in {}. Please run the script with '--help'".format(args.gtJsonFile))
if args.gtFolder is None:
args.gtFolder = os.path.splitext(args.gtJsonFile)[0]
if not os.path.isfile(args.predictionJsonFile):
printError("Could not find a prediction json file in {}. Please run the script with '--help'".format(args.predictionJsonFile))
if args.predictionFolder is None:
args.predictionFolder = os.path.splitext(args.predictionJsonFile)[0]
evaluatePanoptic(args.gtJsonFile, args.gtFolder, args.predictionJsonFile, args.predictionFolder, args.resultsFile)
return
def main():
# # Where to look for Cityscapes
# if 'CITYSCAPES_DATASET' in os.environ:
# cityscapesPath = os.environ['CITYSCAPES_DATASET']
# else:
# cityscapesPath = os.path.join(os.path.dirname(os.path.realpath(__file__)),'..','..')
# # how to search for all ground truth
# searchFine = os.path.join( cityscapesPath , "gtFine" , "*" , "*" , "*_gt*_polygons.json" ) # i. ex: ~~\gtFine\val\frankfurt\~~polygons.json
# searchCoarse = os.path.join( cityscapesPath , "gtCoarse" , "*" , "*" , "*_gt*_polygons.json" )
# # search files
# filesFine = glob.glob( searchFine )
# filesFine.sort()
# filesCoarse = glob.glob( searchCoarse )
# filesCoarse.sort()
# # concatenate fine and coarse
# files = filesFine + filesCoarse
# # files = filesFine # use this line if fine is enough for now.
# Where to look for Cityscapes
# i.21.3.11.12:45) 기존의 convertTestJ 폴더에서 panopticSeg_dentPanoJ 로 폴더명 바꿨고, 그안에 gt 및 inputOriPano 이렇게 두개 폴더 다시 만들어줬음.
# 따라서 ~~polygons.json 경로 바꼈음. 바뀐 ~~polygons.json 경로 ex: panopticSeg_dentPanoJ\gt\train\imp2_1_polygons.json
# MYROOTDIRPATH_J = r"C:\Users\starriet\Downloads\panopticSeg_dentPanoJ" # ~~polygons.json 경로 ex: convertTestJ\train\imp2_1_polygons.json # <-요건 기존경로.
# i.21.3.14.22:41) 코랩컴에서의 경로로 수정. 내 구글드라이브에 커스텀데이터 올려놓고, 코랩컴에서 구글드라이브의 압축파일을 (코랩컴의 디렉토리에다가)압축풀어서 사용할거니까.
# 즉, 구글코랩에서 구글드라이브 연동해서 돌리는걸 가정한것임. 뭐 사실상 코랩에서만 할테니까 일단은.
MYROOTDIRPATH_J = "/content/datasetsJ/panopticSeg_dentPanoJ"
# how to search for all ground truth(i. ~~polygons.json)
forSearchAllPolygonsJson = os.path.join(MYROOTDIRPATH_J, "gt", "*",
"*_polygons.json")
# search files
polygonsjson_path_list = glob.glob(forSearchAllPolygonsJson)
files = polygonsjson_path_list
# quit if we did not find anything
if not files:
printError("j) Did not find any files(~~polygons.json)!!!")
# a bit verbose
print("Processing {} annotation files".format(len(files)))
# iterate through files
progress = 0
print("Progress: {:>3} %".format(progress * 100 / len(files)), end=' ')
for f in files:
# create the output filename
dst = f.replace("_polygons.json", "_labelTrainIds.png")
# do the conversion
try:
json2labelImg(f, dst, "trainIds")
except:
print("Failed to convert: {}".format(f))
raise
# status
progress += 1
print("\rProgress: {:>3} %".format(progress * 100 / len(files)),
end=' ')
sys.stdout.flush()
def convert2panoptic(cityscapesPath=None, outputFolder=None, useTrainId=False):
# Where to look for Cityscapes
if cityscapesPath is None:
if 'CITYSCAPES_DATASET' in os.environ:
cityscapesPath = os.environ['CITYSCAPES_DATASET']
else:
cityscapesPath = os.path.join(os.path.dirname(os.path.realpath(__file__)),'..','..')
cityscapesPath = os.path.join(cityscapesPath, "gtFine")
if outputFolder is None:
outputFolder = cityscapesPath
categories = []
for label in labels:
if label.ignoreInEval:
continue
categories.append({'id': int(label.trainId) if useTrainId else int(label.id),
'name': label.name,
'color': label.color,
'supercategory': label.category,
'isthing': 1 if label.hasInstances else 0})
# if only val set needs the conversion.
# for setName in ["val"]:
for setName in ["val", "train", "test"]:
# how to search for all ground truth
searchFine = os.path.join(cityscapesPath, setName, "*", "*_instanceIds.png")
# search files
filesFine = glob.glob(searchFine)
filesFine.sort()
files = filesFine
# quit if we did not find anything
if not files:
printError(
"Did not find any files for {} set using matching pattern {}. Please consult the README.".format(setName, searchFine)
)
# a bit verbose
print("Converting {} annotation files for {} set.".format(len(files), setName))
trainIfSuffix = "_trainId" if useTrainId else ""
outputBaseFile = "cityscapes_panoptic_{}{}".format(setName, trainIfSuffix)
outFile = os.path.join(outputFolder, "{}.json".format(outputBaseFile))
print("Json file with the annotations in panoptic format will be saved in {}".format(outFile))
panopticFolder = os.path.join(outputFolder, outputBaseFile)
if not os.path.isdir(panopticFolder):
print("Creating folder {} for panoptic segmentation PNGs".format(panopticFolder))
os.mkdir(panopticFolder)
print("Corresponding segmentations in .png format will be saved in {}".format(panopticFolder))
images = []
annotations = []
for progress, f in enumerate(files):
originalFormat = np.array(Image.open(f))
fileName = os.path.basename(f)
imageId = fileName.replace("_gtFine_instanceIds.png", "")
inputFileName = fileName.replace("_instanceIds.png", "_leftImg8bit.png")
outputFileName = fileName.replace("_instanceIds.png", "_panoptic.png")
# image entry, id for image is its filename without extension
images.append({"id": imageId,
"width": int(originalFormat.shape[1]),
"height": int(originalFormat.shape[0]),
"file_name": inputFileName})
pan_format = np.zeros(
(originalFormat.shape[0], originalFormat.shape[1], 3), dtype=np.uint8
)
segmentIds = np.unique(originalFormat)
segmInfo = []
for segmentId in segmentIds:
if segmentId < 1000:
semanticId = segmentId
isCrowd = 1
else:
semanticId = segmentId // 1000
isCrowd = 0
labelInfo = id2label[semanticId]
categoryId = labelInfo.trainId if useTrainId else labelInfo.id
if labelInfo.ignoreInEval:
continue
if not labelInfo.hasInstances:
isCrowd = 0
mask = originalFormat == segmentId
color = [segmentId % 256, segmentId // 256, segmentId // 256 // 256]
pan_format[mask] = color
area = np.sum(mask) # segment area computation
# bbox computation for a segment
hor = np.sum(mask, axis=0)
hor_idx = np.nonzero(hor)[0]
x = hor_idx[0]
width = hor_idx[-1] - x + 1
vert = np.sum(mask, axis=1)
vert_idx = np.nonzero(vert)[0]
y = vert_idx[0]
height = vert_idx[-1] - y + 1
bbox = [int(x), int(y), int(width), int(height)]
segmInfo.append({"id": int(segmentId),
"category_id": int(categoryId),
"area": int(area),
"bbox": bbox,
"iscrowd": isCrowd})
annotations.append({'image_id': imageId,
'file_name': outputFileName,
"segments_info": segmInfo})
Image.fromarray(pan_format).save(os.path.join(panopticFolder, outputFileName))
print("\rProgress: {:>3.2f} %".format((progress + 1) * 100 / len(files)), end=' ')
sys.stdout.flush()
print("\nSaving the json file {}".format(outFile))
d = {'images': images,
'annotations': annotations,
'categories': categories}
with open(outFile, 'w') as f:
json.dump(d, f, sort_keys=True, indent=4)
