def get_packed_output(interpreter, score_threshold, top_k, image_scale=1.0):
boxes = common.output_tensor(interpreter, 0)
class_ids = common.output_tensor(interpreter, 1)
scores = common.output_tensor(interpreter, 2)
count = int(common.output_tensor(interpreter, 3))
top_k_scores_id = np.argpartition(np.array(scores), top_k)[-top_k:]
top_k_scores = scores[top_k_scores_id]
threshold_pass = top_k_scores_id[top_k_scores >= score_threshold]
boxes, scores, class_ids = boxes[threshold_pass], scores[threshold_pass], class_ids[threshold_pass]
return [boxes, scores, class_ids, count]
def get_output(interpreter, score_threshold, top_k, image_scale=1.0):
"""Returns list of detected objects."""
boxes = common.output_tensor(interpreter, 0)
category_ids = common.output_tensor(interpreter, 1)
scores = common.output_tensor(interpreter, 2)
def make(i):
ymin, xmin, ymax, xmax = boxes[i]
return Object(id=int(category_ids[i]),
score=scores[i],
bbox=BBox(xmin=np.maximum(0.0, xmin),
ymin=np.maximum(0.0, ymin),
xmax=np.minimum(1.0, xmax),
ymax=np.minimum(1.0, ymax)))
return [make(i) for i in range(top_k) if scores[i] >= score_threshold]
def get_output(self, interpreter, score_threshold, top_k, image_scale=1.0):
boxes = common.output_tensor(interpreter, 0)
class_ids = common.output_tensor(interpreter, 1)
scores = common.output_tensor(interpreter, 2)
count = int(common.output_tensor(interpreter, 3))
def make(i):
ymin, xmin, ymax, xmax = boxes[i]
return Object(id=int(class_ids[i]),
score=scores[i],
bbox=BBox(xmin=np.maximum(0.0, xmin),
ymin=np.maximum(0.0, ymin),
xmax=np.minimum(1.0, xmax),
ymax=np.minimum(1.0, ymax)))
return [make(i) for i in range(top_k) if scores[i] >= score_threshold]
def get_detection_output(interpreter):
threshold = .6
boxes = common.output_tensor(interpreter, 0)
classes = common.output_tensor(interpreter, 1)
scores = common.output_tensor(interpreter, 2)
count = int(common.output_tensor(interpreter, 3))
hand_detections = []
for i in range(count):
if scores[i] > threshold:
hand_detections.append(boxes[i])
#print(hand_detections)
#sys.stdout.flush()
return hand_detections
def get_emotion(interpreter, top_k=1, score_threshold=0.1):
scores = common.output_tensor(interpreter, 0)
print(scores)
categories = [
Category(i, scores[i])
for i in np.argpartition(scores, -top_k)[-top_k:]
if scores[i] >= score_threshold
]
return sorted(categories, key=operator.itemgetter(1), reverse=True)
def get_output(interpreter, top_k, score_threshold):
"""Returns no more than top_k categories with score >= score_threshold."""
scores = common.output_tensor(interpreter, 0)
categories = [
Category(i, scores[i])
for i in np.argpartition(scores, -top_k)[-top_k:]
if scores[i] >= score_threshold
]
return sorted(categories, key=operator.itemgetter(1), reverse=True)
def get_output(interpreter, image_scale=1.0):
"""Returns list of detected objects."""
boxes = common.output_tensor(interpreter, 0)
class_ids = common.output_tensor(interpreter, 1)
scores = common.output_tensor(interpreter, 2)
count = int(common.output_tensor(interpreter, 3))
def make(i):
ymin, xmin, ymax, xmax = boxes[i]
return Object(id=int(class_ids[i]),
score=scores[i],
bbox=BBox(xmin=np.maximum(0.0, xmin),
ymin=np.maximum(0.0, ymin),
xmax=np.minimum(1.0, xmax),
ymax=np.minimum(1.0, ymax)))
return [make(i) for i in range(len(scores))
if not np.isnan(class_ids[i])] #if scores[i] >= score_threshold]
def get_output_CAM(interpreter, dense_layer_weight, cam_w, cam_h):
scaling = False
CAM = common.output_tensor(interpreter,
0,
dense_layer_weight,
scaling=scaling)
if scaling == True:
pass
else:
CAM = cv2.resize(CAM, (cam_w, cam_h)) # 이거 한 번 하자고 cv2 부르니..?
return CAM
def create_pose(model, points):
common.set_input_pose(model, points)
model.invoke()
scores = common.output_tensor(model, 0)
x = points[:, 0::2]
y = points[:, 1::2]
z_pred = np.array(scores)
z_pred = np.reshape(z_pred, (-1, len(scores)))
pose = np.stack((x, y, z_pred), axis=-1)
pose = np.reshape(pose, (len(points), -1))
return pose
def gen_frames(): # generate frame by frame from camera
#Current regression model is unstable, so we take running averages of sin, cos, and angle to stabilize the values
moving_window = 10
runsin = np.zeros(moving_window)
runcos = np.zeros(moving_window)
runtheta = np.zeros(moving_window)
while True:
ret, frame = cap.read()
if not ret:
break
cv2_im = frame
cv2_im_rgb = cv2.cvtColor(cv2_im, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im_rgb)
common.set_input(interpreter, pil_im)
interpreter.invoke()
#Uncomment the following if you need to use bounding boxes or classification schemes and to use it for overlay
#objs = get_output(interpreter, score_threshold=args.threshold, top_k=args.top_k)
#cv2_im = append_objs_to_img(cv2_im, objs, labels)
sincos = common.output_tensor(interpreter, 0)
runsin = np.roll(runsin, 1)
runsin[0] = sincos[0]
runcos = np.roll(runcos, 1)
runcos[0] = sincos[1]
runtheta = np.roll(runtheta, 1)
runtheta[0] = 180 / np.pi * np.arctan2(sincos[0], sincos[1])
cv2_im = cv2.putText(cv2_im, 'angle: {}'.format(np.average(runtheta)),
(0, 480 - 90), cv2.FONT_HERSHEY_SIMPLEX, 1.0,
(255, 0, 0), 2)
cv2_im = cv2.putText(cv2_im, 'sin: {}'.format(np.average(runsin)),
(0, 480 - 50), cv2.FONT_HERSHEY_SIMPLEX, 1.0,
(255, 0, 0), 2)
cv2_im = cv2.putText(cv2_im, 'cos: {}'.format(np.average(runcos)),
(0, 480 - 10), cv2.FONT_HERSHEY_SIMPLEX, 1.0,
(255, 0, 0), 2)
sd.putNumber(
"WOF Angle",
180 / np.pi * np.arctan2(np.average(runsin), np.average(runcos)))
ret, buffer = cv2.imencode('.jpg', frame)
frame = buffer.tobytes()
yield (b'--frame\r\n'
b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n'
) # concat frame one by one and show result
if cv2.waitKey(1) & 0xFF == ord('q'):
break
def get_features(interpreter, patches):
#patches je lista ndarray objekata, treba pretvoriti u tenzor istih dimenzija ili sliku po sliku ubacivat
features = []
for patch in patches:
if 0 in patch.shape:
features.append(None)
continue
common.set_input(interpreter, Image.fromarray(patch))
interpreter.invoke()
feature = common.output_tensor(interpreter, 0)
features.append(feature)
return features
def get_classification_output(interpreter):
scores = common.output_tensor(interpreter, 0)
#print(scores)
#sys.stdout.flush()
return scores