def __getitem__(self, idx):
epiInd = 0 # calculate the epiInd
while idx >= self.episodeNum[epiInd]:
# print self.episodeNum[epiInd],
epiInd += 1
if epiInd > 0:
idx -= self.episodeNum[epiInd - 1]
# random fliping
flipping = False
if self.aug and random.random() > 0.5:
flipping = True
# print epiInd, idx
imgseq = []
for k in range(self.batch):
img = cv2.imread(self.imgnamelist[epiInd][idx + k])
if self.aug:
img = im_hsv_augmentation(img)
img = im_crop(img)
outimg = im_scale_norm_pad(img,
outsize=self.imgsize,
mean=self.mean,
std=self.std,
down_reso=True,
flip=flipping)
imgseq.append(outimg)
return np.array(imgseq)
def detect(image, confidence_thres=0.8, display_label=True):
### detect and mark a person bb in image, return the marked image
(h, w) = image.shape[:2]
# print(h, w)
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843,
(300, 300), 127.5)
net.setInput(blob)
detections = net.forward()
for i in np.arange(0, detections.shape[2]):
confidence = detections[0, 0, i, 2]
if confidence > confidence_thres:
idx = int(detections[0, 0, i, 1])
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
if CLASSES[idx] == "person":
mX = (endX - startX) // 10
mY = (endY - startY) // 10
startX = max(startX - mX, 0)
endX = min(endX + mX, w - 1)
startY = max(startY - mY, 0)
endY = min(endY + mY, h - 1)
# get direction
subimg = image[startY:endY, startX:endX]
subimg = utils.im_scale_norm_pad(subimg,
mean=[127.5, 127.5, 127.5],
std=[127.5, 127.5, 127.5])
inp = torch.tensor(np.array([subimg]))
angle = predictor.forward(inp)
angle = angle.data[0]
alpha = np.array(angle)
# display bounding box
out_img = utils.img_denormalize(subimg, [127.5, 127.5, 127.5],
[127.5, 127.5, 127.5])
put_arrow(out_img, alpha)
out_img = cv2.resize(out_img, (0, 0), fx=3.0, fy=3.0)
cv2.imshow("Frame", out_img)
key = cv2.waitKey(5)
cv2.rectangle(image, (startX, startY), (endX, endY),
COLORS[idx], 2)
return image
def augment_image(self, img, flipping):
# augment image to make "new" data
if self.aug:
img = utils.im_hsv_augmentation(img)
img = utils.im_crop(img, maxscale=self.maxscale)
out_img = utils.im_scale_norm_pad(
img,
self.mean,
self.std,
out_size=self.img_size,
# down_reso=True,
flip=flipping)
return out_img
def __getitem__(self, idx):
img = cv2.imread(self.imgnamelist[idx]) # in bgr
label = np.array(self.labellist[idx], dtype=np.float32)
# random fliping
flipping = False
if self.aug and random.random() > 0.5:
flipping = True
label[1] = -label[1]
if self.aug:
img = im_hsv_augmentation(img)
img = im_crop(img, maxscale=self.maxscale)
outimg = im_scale_norm_pad(img,
outsize=self.imgsize,
mean=self.mean,
std=self.std,
down_reso=True,
flip=flipping)
return {'img': outimg, 'label': label}
def __getitem__(self, idx):
epiInd = 0 # calculate the epiInd
while idx >= self.episodeNum[epiInd]:
# print self.episodeNum[epiInd],
epiInd += 1
if epiInd > 0:
idx -= self.episodeNum[epiInd - 1]
# random fliping
flipping = False
if self.aug and random.random() > 0.5:
flipping = True
# print epiInd, idx
imgseq = []
labelseq = []
for k in range(self.batch):
img = cv2.imread(self.imgnamelist[epiInd][idx + k][0])
angle = self.imgnamelist[epiInd][idx + k][1]
direction_angle_cos = np.cos(float(angle))
direction_angle_sin = np.sin(float(angle))
label = np.array([direction_angle_sin, direction_angle_cos],
dtype=np.float32)
if self.aug:
img = im_hsv_augmentation(img)
img = im_crop(img)
if flipping:
label[1] = -label[1]
outimg = im_scale_norm_pad(img,
outsize=self.imgsize,
mean=self.mean,
std=self.std,
down_reso=True,
flip=flipping)
imgseq.append(outimg)
labelseq.append(label)
return {'imgseq': np.array(imgseq), 'labelseq': np.array(labelseq)}
def __getitem__(self, idx):
if self.filename[-3:] == 'csv':
point_info = self.items.iloc[idx]
else:
point_info = self.items[idx]
#print(point_info)
img_name = point_info['path']
direction_angle = point_info['direction_angle']
direction_angle_cos = np.cos(float(direction_angle))
direction_angle_sin = np.sin(float(direction_angle))
label = np.array([direction_angle_sin, direction_angle_cos],
dtype=np.float32)
img = cv2.imread(img_name)
# random fliping
flipping = False
if self.aug and random.random() > 0.5:
flipping = True
label[1] = -label[1]
if img is None:
print 'error image:', img_name
return
if self.aug:
img = im_hsv_augmentation(img, Hscale=10, Sscale=60, Vscale=60)
img = im_crop(img, maxscale=self.maxscale)
outimg = im_scale_norm_pad(img,
outsize=self.imgsize,
mean=self.mean,
std=self.std,
down_reso=True,
flip=flipping)
return {'img': outimg, 'label': label}