def assign_tags(video_id):
import django
from PIL import Image
sys.path.append(os.path.dirname(__file__))
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "dva.settings")
django.setup()
from django.conf import settings
from dvaapp.models import Video, Frame, Region
from dvalib import entity, annotator
dv = Video.objects.get(id=video_id)
frames = Frame.objects.all().filter(video=dv)
v = entity.WVideo(dvideo=dv, media_dir=settings.MEDIA_ROOT)
wframes = {
df.pk: entity.WFrame(video=v,
frame_index=df.frame_index,
primary_key=df.pk)
for df in frames
}
algorithm = annotator.OpenImagesAnnotator()
logging.info("starting annotation {}".format(algorithm.name))
for k, f in wframes.items():
tags = algorithm.apply(f.local_path())
a = Region()
a.region_type = Region.ANNOTATION
a.frame_id = k
a.video_id = video_id
a.object_name = "OpenImagesTag"
a.metadata_text = " ".join([t for t, v in tags.iteritems() if v > 0.1])
a.metadata_json = json.dumps(
{t: 100.0 * v
for t, v in tags.iteritems() if v > 0.1})
a.full_frame = True
a.save()
print a.metadata_text
def perform_detection(video_id):
start = TEvent()
start.video_id = video_id
start.started = True
start.operation = "detection"
start.save()
start_time = time.time()
dv = Video.objects.get(id=video_id)
frames = Frame.objects.all().filter(video=dv)
v = entity.WVideo(dvideo=dv, media_dir=settings.MEDIA_ROOT)
wframes = [entity.WFrame(video=v, frame_index=df.frame_index, primary_key=df.pk) for df in frames]
darknet_path = os.path.join(settings.BASE_DIR,'darknet/')
list_path = "{}/{}_list.txt".format(darknet_path,os.getpid())
output_path = "{}/{}_output.txt".format(darknet_path,os.getpid())
logging.info(darknet_path)
path_to_pk = {}
with open(list_path,'w') as framelist:
for frame in wframes:
framelist.write('{}\n'.format(frame.local_path()))
path_to_pk[frame.local_path()] = frame.primary_key
#./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg yolo9000.weights data/list.txt
with open(output_path,'w') as output:
args = ["./darknet", 'detector', 'test', 'cfg/combine9k.data', 'cfg/yolo9000.cfg', 'yolo9000.weights', list_path]
logging.info(args)
returncode = subprocess.call(args,cwd=darknet_path,stdout=output)
if returncode == 0:
detections = 0
for line in file(output_path):
if line.strip():
detections += 1
frame_path,name,confidence,left,right,top,bot = line.strip().split('\t')
if frame_path not in path_to_pk:
raise ValueError,frame_path
top = int(top)
left = int(left)
right = int(right)
bot = int(bot)
confidence = float(confidence)
dd = Detection()
dd.video = dv
dd.frame_id = path_to_pk[frame_path]
dd.object_name = "darknet_yolo9k_{}".format(name.replace(' ','_'))
dd.confidence = confidence
dd.x = left
dd.y = top
dd.w = right - left
dd.h = bot - top
dd.save()
img = Image.open(frame_path)
img2 = img.crop((left, top, right,bot))
img2.save("{}/{}/detections/{}.jpg".format(settings.MEDIA_ROOT,video_id,dd.pk))
dv.detections = detections
dv.save()
start.completed = True
start.seconds = time.time() - start_time
start.save()
return returncode
def detect(video_id):
"""
This is a HACK since Tensorflow is absolutely atrocious in allocating and freeing up memory.
Once a process / session is allocated a memory it cannot be forced to clear it up.
As a result this code gets called via a subprocess which clears memory when it exits.
:param video_id:
:return:
"""
import django
from PIL import Image
sys.path.append(os.path.dirname(__file__))
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "dva.settings")
django.setup()
from django.conf import settings
from dvaapp.models import Video, Detection, Frame
from dvalib import entity, detector
dv = Video.objects.get(id=video_id)
frames = Frame.objects.all().filter(video=dv)
v = entity.WVideo(dvideo=dv, media_dir=settings.MEDIA_ROOT)
wframes = {
df.pk: entity.WFrame(video=v,
frame_index=df.frame_index,
primary_key=df.pk)
for df in frames
}
detection_count = 0
detector_list = {
'ssd': detector.SSDetector(),
}
if 'YOLO_ENABLE' in os.environ:
detector_list['yolo'] = detector.YOLODetector()
for alogrithm in detector_list.itervalues():
logging.info("starting detection {}".format(alogrithm.name))
frame_detections = alogrithm.detect(wframes.values())
for frame_pk, detections in frame_detections.iteritems():
for d in detections:
dd = Detection()
dd.video = dv
dd.frame_id = frame_pk
dd.object_name = d['name']
dd.confidence = d['confidence']
dd.x = d['left']
dd.y = d['top']
dd.w = d['right'] - d['left']
dd.h = d['bot'] - d['top']
dd.save()
img = Image.open(wframes[frame_pk].local_path())
img2 = img.crop((d['left'], d['top'], d['right'], d['bot']))
img2.save("{}/{}/detections/{}.jpg".format(
settings.MEDIA_ROOT, video_id, dd.pk))
detection_count += 1
dv.refresh_from_db()
dv.detections = dv.detections + detection_count
dv.save()
def perform_face_indexing(video_id):
face_indexer = indexer.FacenetIndexer()
dv = Video.objects.get(id=video_id)
video = entity.WVideo(dv, settings.MEDIA_ROOT)
frames = Frame.objects.all().filter(video=dv)
wframes = [
entity.WFrame(video=video,
frame_index=df.frame_index,
primary_key=df.pk) for df in frames
]
input_paths = {f.local_path(): f.primary_key for f in wframes}
faces_dir = '{}/{}/detections'.format(settings.MEDIA_ROOT, video_id)
indexes_dir = '{}/{}/indexes'.format(settings.MEDIA_ROOT, video_id)
face_detector = detector.FaceDetector()
aligned_paths = face_detector.detect(wframes)
logging.info(len(aligned_paths))
faces = []
faces_to_pk = {}
count = 0
for path, v in aligned_paths.iteritems():
for scaled_img, bb in v:
d = Region()
d.region_type = Region.DETECTION
d.video = dv
d.confidence = 100.0
d.frame_id = input_paths[path]
d.object_name = "mtcnn_face"
left, top, right, bottom = bb[0], bb[1], bb[2], bb[3]
d.y = top
d.x = left
d.w = right - left
d.h = bottom - top
d.save()
face_path = '{}/{}.jpg'.format(faces_dir, d.pk)
output_filename = os.path.join(faces_dir, face_path)
misc.imsave(output_filename, scaled_img)
faces.append(face_path)
faces_to_pk[face_path] = d.pk
count += 1
dv.refresh_from_db()
dv.detections = dv.detections + count
dv.save()
path_count, emb_array, entries, feat_fname, entries_fname = face_indexer.index_faces(
faces, faces_to_pk, indexes_dir, video_id)
i = IndexEntries()
i.video = dv
i.count = len(entries)
i.contains_frames = False
i.contains_detections = True
i.detection_name = "Face"
i.algorithm = 'facenet'
i.entries_file_name = entries_fname.split('/')[-1]
i.features_file_name = feat_fname.split('/')[-1]
i.save()
def perform_face_indexing(video_id):
dv = Video.objects.get(id=video_id)
video = entity.WVideo(dv, settings.MEDIA_ROOT)
frames = Frame.objects.all().filter(video=dv)
wframes = [
entity.WFrame(video=video,
frame_index=df.frame_index,
primary_key=df.pk) for df in frames
]
input_paths = {f.local_path(): f.primary_key for f in wframes}
faces_dir = '{}/{}/detections'.format(settings.MEDIA_ROOT, video_id)
indexes_dir = '{}/{}/indexes'.format(settings.MEDIA_ROOT, video_id)
aligned_paths = facerecognition.align(input_paths.keys(), faces_dir)
logging.info(len(aligned_paths))
faces = []
faces_to_pk = {}
count = 0
for path, v in aligned_paths.iteritems():
for face_path, bb in v:
d = Detection()
d.video = dv
d.confidence = 100.0
d.frame_id = input_paths[path]
d.object_name = "mtcnn_face"
top, left, bottom, right = bb[0], bb[1], bb[2], bb[3]
d.y = top
d.x = left
d.w = right - left
d.h = bottom - top
d.save()
os.rename(face_path, '{}/{}.jpg'.format(faces_dir, d.pk))
faces.append('{}/{}.jpg'.format(faces_dir, d.pk))
faces_to_pk['{}/{}.jpg'.format(faces_dir, d.pk)] = d.pk
count += 1
dv.detections = dv.detections + count
dv.save()
path_count, emb_array, entries = facerecognition.represent(
faces, faces_to_pk, indexes_dir)
i = IndexEntries()
i.video = dv
i.count = len(entries)
i.contains_frames = False
i.contains_detections = True
i.detection_name = "Face"
i.algorithm = 'facenet'
i.save()
def detect(video_id):
"""
This is a HACK since Tensorflow is absolutely atrocious in allocating and freeing up memory
:param video_id:
:return:
"""
import django
from PIL import Image
sys.path.append(os.path.dirname(__file__))
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "dva.settings")
django.setup()
from django.conf import settings
from dvaapp.models import Video, Detection, Frame
from dvalib import entity, detector
dv = Video.objects.get(id=video_id)
frames = Frame.objects.all().filter(video=dv)
v = entity.WVideo(dvideo=dv, media_dir=settings.MEDIA_ROOT)
wframes = {
df.pk: entity.WFrame(video=v,
frame_index=df.frame_index,
primary_key=df.pk)
for df in frames
}
detection_count = 0
for alogrithm in detector.DETECTORS.itervalues():
logging.info("starting detection {}".format(alogrithm.name))
frame_detections = alogrithm.detect(wframes.values())
for frame_pk, detections in frame_detections.iteritems():
for d in detections:
dd = Detection()
dd.video = dv
dd.frame_id = frame_pk
dd.object_name = d['name']
dd.confidence = d['confidence']
dd.x = d['left']
dd.y = d['top']
dd.w = d['right'] - d['left']
dd.h = d['bot'] - d['top']
dd.save()
img = Image.open(wframes[frame_pk].local_path())
img2 = img.crop((d['left'], d['top'], d['right'], d['bot']))
img2.save("{}/{}/detections/{}.jpg".format(
settings.MEDIA_ROOT, video_id, dd.pk))
detection_count += 1
dv.detections = detection_count
dv.save()
