def kitti_dataset_analysis(self):
searchlabel = os.path.join(path_kitti, "image_*", "*.png")
fileslabel = glob.glob(searchlabel)
fileslabel.sort()
dilatation = 10
threshold = 0.1
for i in range(len(fileslabel)):
img = cv2.imread(fileslabel[i])
img = cv2.imread(fileslabel[i], -1)
img = img.astype(np.uint16)
img = cv2.resize(img, (1024, 320))
results = self.model.detect([img],
dilatation,
threshold,
verbose=1)
r = results[0]
selected_class = self.class_selection(r['masks'], r['class_ids'])
#selected_class = r['masks']
result_image = visualize_cv.cv_img_masked(img, r['rois'],
selected_class,
r['class_ids'],
self.class_names,
r['scores'])
cv2.imwrite(fileslabel[i], result_image.astype('uint8'))
print("Batch done")
def live_depth_analysis(self):
#Live stream video analysis
while not rospy.is_shutdown():
#print("Looking for a frame")
current_frame = self.frame
current_depth_frame = self.depth_frame
if (current_frame != [] and current_depth_frame != []):
#print("got frame")
#image = cv2.imread("/home/pmcrivet/catkin_ws/src/Mask_RCNN/script/images/frame0057.jpg")
#img = numpy.asarray(frame,dtype='uint8')
#cv2.imshow("image",image)
#cv2.waitKey(0)
#cv2.destroyAllWindows()
#print image.shape
#print image.dtype
#Dilatation effect on masks for better feature masking
dilatation = 20
threshold = 0.3
results = self.model.detect([current_frame],
dilatation,
threshold,
verbose=1)
r = results[0]
#selected_class = self.class_selection(r['masks'], r['class_ids'])
selected_class = r['masks']
result_image = visualize_cv.cv_img_masked(
current_frame, r['rois'], selected_class, r['class_ids'],
self.class_names, r['scores'])
result_depth_image = visualize_cv.cv_depth_img_masked(
current_depth_frame, r['rois'], selected_class,
r['class_ids'], self.class_names, r['scores'])
DITS = Image()
ITS = Image()
DITS = self.bridge.cv2_to_imgmsg(result_depth_image, '16UC1')
ITS = self.bridge.cv2_to_imgmsg(result_image, 'bgr8')
DITS.header = self.depth_msg_header
ITS.header = self.msg_header
self.image_pub.publish(ITS)
self.depth_image_pub.publish(DITS)
print("Publishing img")
else:
print("Waiting for frame")
def process_image(self, image):
dilatation = 1
threshold = 0.1
results = self.model.detect([image], dilatation, threshold, verbose=1)
r = results[0]
selected_class = r['masks']
result_image = visualize_cv.cv_img_masked(image, r['rois'],
selected_class,
r['class_ids'],
self.class_names,
r['scores'])
return result_image
def live_analysis(self):
#Live stream video analysis
while not rospy.is_shutdown():
#print("Looking for a frame")
current_frame = self.frame
if current_frame != []:
#print("got frame")
#image = cv2.imread("/home/pmcrivet/catkin_ws/src/Mask_RCNN/script/images/frame0057.jpg")
#img = numpy.asarray(frame,dtype='uint8')
#cv2.imshow("image",image)
#cv2.waitKey(0)
#cv2.destroyAllWindows()
#print image.shape
#print image.dtype
results = self.model.detect([current_frame], verbose=1)
r = results[0]
result_image = visualize_cv.cv_img_masked(
current_frame, r['rois'], r['masks'], r['class_ids'],
self.class_names, r['scores'])
cv2.namedWindow('result_image', cv2.WINDOW_NORMAL)
#cv2.resizeWindow('result_image', 1920, 1080)
cv2.imshow("result_image", result_image)
print("Did something")
#results = self.model.detect([self.frame], verbose=1)
#r = results[0]
#visualize_cv.display_instances(image, r['rois'], r['masks'], r['class_ids'],self.class_names, r['scores'])
ITS = Image()
ITS = self.bridge.cv2_to_imgmsg(result_image, 'bgr8')
ITS.header = self.msg_header
self.image_pub.publish(ITS)
print("Publishing img")
else:
print("Waiting for frame")
def tum_dataset_analysis(self):
searchlabel = os.path.join(path_tum, "rgb_sync", "*.png")
fileslabel = glob.glob(searchlabel)
fileslabel.sort()
searchanot = os.path.join(path_tum, "depth_sync", "*.png")
filesanot = glob.glob(searchanot)
filesanot.sort()
dilatation = 20
threshold = 0.3
for i in range(len(filesanot)):
img = cv2.imread(fileslabel[i])
#depth_img = cv2.cvtColor(cv2.imread(filesanot[i]), cv2.COLOR_BGR2GRAY)
depth_img = cv2.imread(filesanot[i], -1)
depth_img = depth_img.astype(np.uint16)
results = self.model.detect([img],
dilatation,
threshold,
verbose=1)
r = results[0]
selected_class = self.class_selection(r['masks'], r['class_ids'])
#selected_class = r['masks']
result_image = visualize_cv.cv_img_masked(img, r['rois'],
selected_class,
r['class_ids'],
self.class_names,
r['scores'])
result_depth_image = visualize_cv.cv_depth_img_masked(
depth_img, r['rois'], selected_class, r['class_ids'],
self.class_names, r['scores'])
cv2.imwrite(filesanot[i], result_depth_image)
print("Batch done")