def copy_frames(self, demonstration, frm, l1_cluster, l2_cluster, surgeme):
from_path = (
constants.PATH_TO_DATA
+ constants.NEW_FRAMES_FOLDER
+ demonstration
+ "_"
+ constants.CAMERA
+ "/"
+ utils.get_frame_fig_name(frm)
)
to_path = (
constants.PATH_TO_CLUSTERING_RESULTS
+ self.trial
+ "/"
+ l1_cluster
+ "/"
+ l2_cluster
+ "_"
+ str(surgeme)
+ "_"
+ demonstration
+ "_"
+ utils.get_frame_fig_name(frm)
)
utils.sys_copy(from_path, to_path)
def run_video_with_bsub(cap, func, kernel = None, params = None):
if params is not None:
fgbg = func(params[0], params[1], params[2], params[3])
else:
fgbg = func()
SAVE_PATH = constants.PATH_TO_DATA + constants.NEW_BGSUB_FOLDER + "Suturing_E003_capture2/"
i = 1
while(1):
ret, frame = cap.read()
print i
fgmask = fgbg.apply(frame)
mask_rbg = cv2.cvtColor(fgmask, cv2.COLOR_GRAY2BGR)
draw = frame & mask_rbg
if frame == None and mask_rbg == None:
sys.exit()
if kernel is not None:
fgmask = cv2.morphologyEx(fgmask, cv2.MORPH_OPEN, kernel)
# cv2.namedWindow("frame", cv2.WND_PROP_FULLSCREEN)
# cv2.setWindowProperty("frame", cv2.WND_PROP_FULLSCREEN, cv2.cv.CV_WINDOW_FULLSCREEN)
cv2.imshow('frame', draw)
cv2.imwrite(SAVE_PATH + utils.get_frame_fig_name(i), draw)
k = cv2.waitKey(30) & 0xff
i += 1
if k == 27:
break
cap.release()
cv2.destroyAllWindows()
def copy_frames(self, demonstration, frm, l1_cluster, l2_cluster, surgeme):
from_path = constants.PATH_TO_DATA + constants.NEW_FRAMES_FOLDER + demonstration + "_" + constants.CAMERA + "/" + utils.get_frame_fig_name(
frm)
to_path = constants.PATH_TO_CLUSTERING_RESULTS + self.trial + "/" + l1_cluster + "/" + l2_cluster + "_" + str(
surgeme) + "_" + demonstration + "_" + utils.get_frame_fig_name(
frm)
utils.sys_copy(from_path, to_path)
def copy_milestone_frames(self, matrix, list_of_cp_key, gmm): neigh = neighbors.KNeighborsClassifier(n_neighbors = 1) neigh.fit(matrix, list_of_cp_key) milestone_closest_cp = neigh.predict(gmm.means_) assert len(milestone_closest_cp) == self.n_components_L2 for cp in milestone_closest_cp: demonstration = self.map_cp2demonstrations[cp] surgeme = self.map_frm2surgeme[demonstration][self.map_cp2frm[cp]] frm = utils.get_frame_fig_name(self.map_cp2frm[cp]) from_path = constants.PATH_TO_DATA + constants.NEW_FRAMES_FOLDER + demonstration +"_" + constants.CAMERA + "/" + frm to_path = constants.PATH_TO_CLUSTERING_RESULTS + self.trial + "/milestones/" + self.map_cp2milestones[cp] + "_" + str(surgeme) + "_" + demonstration + "_" + frm utils.sys_copy(from_path, to_path)
def copy_milestone_frames(self, matrix, list_of_cp_key, gmm): neigh = neighbors.KNeighborsClassifier(n_neighbors = 1) neigh.fit(matrix, list_of_cp_key) milestone_closest_cp = neigh.predict(gmm.means_) assert len(milestone_closest_cp) == 3 for cp in milestone_closest_cp: demonstration = self.map_cp2demonstrations[cp] surgeme = self.map_frm2surgeme[demonstration][self.map_cp2frm[cp]] frm = utils.get_frame_fig_name(self.map_cp2frm[cp]) from_path = constants.PATH_TO_DATA + constants.NEW_FRAMES_FOLDER + demonstration +"_capture2/" + frm to_path = constants.PATH_TO_CLUSTERING_RESULTS + self.trial + "/milestones/" + self.map_cp2milestones[cp] + "_" + str(surgeme) + "_" + demonstration + "_" + frm utils.sys_copy(from_path, to_path)
def generate_train_val_test_files(list_of_videos):
list_of_data = []
for video in list_of_videos:
for camera in ["capture1", "capture2"]:
os.chdir(constants.PATH_TO_DATA)
transcriptions_file = TRANSCRIPTIONS_FOLDER + video + ".txt"
with open(transcriptions_file, "rb") as f:
for line in f:
line = line.split()
start = int(line[0])
end = int(line[1])
surgeme = line[2]
label = constants.map_surgeme_label[surgeme]
i = start
while i <= end:
data = constants.PATH_TO_DATA + constants.NEW_FRAMES_FOLDER + video + "_" + camera + "/" + utils.get_frame_fig_name(
i) + " " + str(label) + " \n"
list_of_data.append(data)
i += 1
random.shuffle(list_of_data)
N = len(list_of_data)
train, test = train_test_split(list_of_data,
test_size=constants.train_test_ratio)
val = test[:len(test) / 2]
test = test[len(test) / 2:]
train_file_name = constants.PATH_TO_DATA + "train.txt"
val_file_name = constants.PATH_TO_DATA + "val.txt"
test_file_name = constants.PATH_TO_DATA + "test.txt"
IPython.embed()
write_to_file(train_file_name, train)
write_to_file(val_file_name, val)
write_to_file(test_file_name, test)
def generate_train_val_test_files(list_of_videos): list_of_data = [] for video in list_of_videos: for camera in ["capture1", "capture2"]: os.chdir(constants.PATH_TO_DATA) transcriptions_file = TRANSCRIPTIONS_FOLDER + video + ".txt" with open(transcriptions_file, "rb") as f: for line in f: line = line.split() start = int(line[0]) end = int(line[1]) surgeme = line[2] label = constants.map_surgeme_label[surgeme] i = start while i <= end: data = constants.PATH_TO_DATA + constants.NEW_FRAMES_FOLDER + video + "_" + camera + "/" + utils.get_frame_fig_name(i) + " "+ str(label) + " \n" list_of_data.append(data) i += 1 random.shuffle(list_of_data) N = len(list_of_data) train, test = train_test_split(list_of_data, test_size = constants.train_test_ratio) val = test[:len(test)/2] test = test[len(test)/2:] train_file_name = constants.PATH_TO_DATA + "train.txt" val_file_name = constants.PATH_TO_DATA + "val.txt" test_file_name = constants.PATH_TO_DATA + "test.txt" IPython.embed() write_to_file(train_file_name, train) write_to_file(val_file_name, val) write_to_file(test_file_name, test)
