def make_containers(vid_path):
dssp = DatasetSourceProvider()
dssp.add_folder(folder_path=vid_path, dataset_name='playtime')
dssp_tracks = DatasetSourceProvider()
dssp_tracks.add_folder(folder_path=vid_path, dataset_name='tracks')
container = AnnotationContainer(dataset_source_provider=dssp,
dataset_version='0.0.1')
container_tracks = AnnotationContainer(dataset_source_provider=dssp_tracks,
dataset_version='0.0.1')
return container, container_tracks
def detect_on_preview_dir(preview_data, detector):
dsp = DatasetSourceProvider()
dsp.add_source(preview_data, 'predicted')
container = AnnotationContainer(dataset_source_provider=dsp)
for image_path in Path(preview_data).glob('preview*'):
image = load_preview_image(str(image_path))
det = detector.detect_image(image)
e = AnnotationEntry(image_path.name,
ImageSize.from_image(image),
'predicted',
instances=det)
container.add_entry(e)
return container
def main(args):
thresh = .01 if args.box_blending else args.threshold
get_detector = _get_detector_func(args.model)
if get_detector == caffe_detector:
detector = get_detector(args.model,
thresh,
box_blending=args.box_blending,
anchors=args.anchors)
else:
detector = _get_detector_func(args.model)(
args.model, thresh, box_blending=args.box_blending)
with detector:
if args.container:
gt = AnnotationContainer.from_file(args.container)
pred = detector.detect_on_annotation_container(
gt) # .with_selected_labels(['person'])
if args.box_blending and type(detector) == DarknetObjectDetector:
pred = _blend_boxes(pred)
pred.filter_all_instances_by_threshold(args.threshold,
in_place=True)
pred.as_evaluated_against(gt).summary()
elif args.images:
pred = detector.detect_on_image_folder(args.images,
dataset_name='predicted')
if args.box_blending and type(detector) == DarknetObjectDetector:
pred = _blend_boxes(pred)
pred.filter_all_instances_by_threshold(args.threshold,
in_place=True)
elif args.preview:
pred = detect_on_preview_dir(args.preview, detector)
if args.box_blending and type(detector) == DarknetObjectDetector:
pred = _blend_boxes(pred)
pred.filter_all_instances_by_threshold(args.threshold,
in_place=True)
if args.output:
pred.to_file(args.output)
labels = args.labels
if args.container == args.out_container:
if input('Are you sure you want to overwrite container? (y/n): ') != 'y':
quit()
print('Selected labels:')
print(labels)
print('Remove empty entries from trimmed container:', args.remove_empty)
print('Clean container of entries with non-valid paths to images:', args.clean)
if args.out_container is None:
out = 'trimmed_container_{}.bbox'.format('_'.join(args.labels))
else:
out = args.out_container
container = AnnotationContainer.from_file(args.container)
container.with_selected_labels(labels, in_place=True)
if args.remove_empty:
for e in list(container.entries.keys()):
if len(container.entries[e]) == 0:
del container.entries[e]
if args.clean:
for e in list(container.entries.keys()):
if container.entries[e].get_image_full_path() is None or not Path(container.entries[e].get_image_full_path()).exists():
del container.entries[e]
container.summary()
container.to_file(out)
for c in args.container:
assert Path(c).exists(), f'{c} does not exists'
return args
if __name__ == '__main__':
args = get_args()
labels = args.labels
fps = args.fps
colors = defaultdict(lambda: np.random.randint(0, 255, 3))
output_files = []
for container_path in tqdm(args.container, desc='containers'):
name = Path(container_path).stem
container = AnnotationContainer.from_file(container_path)
container.label_to_colour = colors
video_writer = None
output_file_name = Path(
container_path).parent / f'{Path(container_path).stem}.avi'
output_files.append(output_file_name)
if len(labels) > 0:
container.with_selected_labels(labels,
prune_empty_entries=False,
in_place=True)
keys = list(
sorted(container.entries.keys(), key=lambda x: int(Path(x).stem)))
for k in tqdm(keys, desc='frames'):
success, image = vidcap.read()
if success: # and frame < 500:
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
yield frame, image
frame += 1
else:
break
from matplotlib import pyplot as plt
input_side = 64
facenet = cv2.dnn.readNetFromCaffe(
'../demo/face_detector/deploy.prototxt',
'../demo/face_detector/res10_300x300_ssd_iter_140000.caffemodel')
orientation_model = get_orientation_model(input_side=input_side)
container = AnnotationContainer.from_file(
'/home/martin/Desktop/meeting_cut/spaces/no_crop/task_nr_354.bbox')
for fnr, frame in tqdm(get_frame_generator(
'/home/martin/Desktop/meeting_cut/spaces/no_crop/overview.mp4'),
total=5483):
entry = container.entries[str(fnr)]
add_face_and_head_orientation(entry, frame, facenet, orientation_model)
# img = entry.overlaid_on_image(frame)
# plt.imshow(img)
# plt.show()
container.to_file(
'/home/martin/Desktop/meeting_cut/spaces/no_crop/task_nr_354_face_and_orientation.bbox'
)
'-o',
'--output',
type=str,
help='Output filename. If not set, user will be ' +
'prompted by request for output filename during runtime.')
args = parser.parse_args()
if args.output is None:
output_file_name = input('Write file to: ')
else:
output_file_name = args.output
cnt = None
for to_merge in args.containers:
print(f'Merging container {to_merge} ...'.ljust(70), end='')
try:
new_cont = AnnotationContainer.from_file(to_merge)
if cnt is None:
cnt = new_cont
else:
cnt.merge(new_cont,
merge_instance_iou=args.nms,
destructively=True)
del new_cont
gc.collect()
except Exception as e:
print('\tFailed!')
print(e)
continue
print('\tOK')
cnt.to_file(output_file_name)
import sys
from pathlib import Path
from bbox import AnnotationContainer
ALIAS_NAME = 'contsum'
if __name__ == '__main__':
if len(sys.argv) == 1:
print(
'use with arguments <full path to container> or $PWD <container>')
quit()
container = sys.argv[1]
suffix = container.split('.')[-1]
if suffix != 'bbox' and suffix != 'json':
if len(sys.argv) < 2:
print(
'if not using 2 arguments, first must be absolute path to container'
)
quit()
container = Path(container) / sys.argv[2]
AnnotationContainer.from_file(container).summary()
action='store_true',
help='Prune empty entries')
parser.add_argument('-max',
'--max_labels',
type=int,
default=None,
help='Max amount of labels in image')
parser.add_argument('-min',
'--min_labels',
type=int,
default=None,
help='Min amount of labels in image')
args = parser.parse_args()
container = args.container
cont = AnnotationContainer.from_file(container)
new_cont = AnnotationContainer(
dataset_source_provider=cont.dataset_source_provider)
if args.labels and len(args.labels) > 0:
cont = cont.with_selected_labels(args.labels,
prune_empty_entries=args.prune,
in_place=True)
entry_keys = list(cont.entries.keys())
shuffle(entry_keys)
key = None
i = 0
not_found_counter = 0
found_counter = 0
successfully_shown = set()
'--shuffle',
action='store_true',
help='Shuffle images')
parser.add_argument('-t',
'--threshold',
type=float,
help='Filter detections with confidence')
parser.add_argument('-m',
'--max',
type=int,
default=None,
help='Show only this amount of images')
args = parser.parse_args()
container = args.container
cont = AnnotationContainer.from_file(container)
if args.labels and len(args.labels) > 0:
cont = cont.with_selected_labels(args.labels,
prune_empty_entries=args.prune,
in_place=True)
if args.threshold is not None:
assert 0 <= args.threshold <= 1, 'Confidence threshold must be in [0, 1]'
cont = cont.filter_all_instances_by_threshold(args.threshold,
in_place=True)
if args.name:
for n in args.name:
if n not in cont:
print(n, 'not in container')
print('Num instances:', len(cont[n].instances))
cont[n].show()