def walk_current_folder(root_path, mars_opts):
trials_to_run = []
fullpaths = []
for path, subdirs, filenames in os.walk(root_path):
for numv, fname in enumerate(filenames):
# start by looping over all movie files that have "Top" in their name
cond1 = all(x not in fname
for x in ['.seq', '.avi', '.mpg', '.mp4'])
cond2 = 'skipped' in path
cond3 = 'Top' not in fname and '_t.seq' not in fname
if cond1 | cond2 | cond3:
continue
if 'Top' in fname or '_t.seq' in fname:
front_fname, top_fname, mouse_name = mof.get_names(fname)
print(front_fname + ' ' + top_fname + ' ' + mouse_name)
if top_fname != fname: continue
fullpath_to_front = os.path.join(path, front_fname)
fullpath_to_top = os.path.join(path, top_fname)
cond1 = not (os.path.exists(fullpath_to_front))
cond2 = mars_opts['doFront']
cond3 = not mars_opts['doTop']
cond4 = not mars_opts['doToppcf']
# If the front video doesn't exist and we need it, continue.
if (cond1 and cond2): # |(cond3 and cond4):
continue
elif 'Front' in fname:
front_fname = fname
front_fname, top_fname, mouse_name = mof.get_names(front_fname)
# if front_fname != fname: continue
fullpath_to_top = os.path.join(path, top_fname)
fullpath_to_front = os.path.join(path, front_fname)
cond1 = not (os.path.exists(fullpath_to_top))
cond2 = mars_opts['doTop']
cond3 = not mars_opts['doFront']
cond4 = not mars_opts['doToppcf']
if (cond1 and cond2 and cond4) | cond3:
continue
else:
# This is a movie file, but doesnt have "Top" or "Front" in it. Let's skip it.
continue
# Save the paths we want to use.
mouse_trial = dict()
mouse_trial['top'] = fullpath_to_top
mouse_trial['front'] = fullpath_to_front
trials_to_run.append(mouse_trial)
fullpaths.append(fullpath_to_top)
return trials_to_run, fullpaths
def extract_pose_wrapper(video_fullpath,
view,
doOverwrite,
progress_bar_signal='',
verbose=0,
output_suffix='',
mars_opts={},
max_frames=999999):
video_path = os.path.dirname(video_fullpath)
video_name = os.path.basename(video_fullpath)
output_folder = mof.get_mouse_output_dir(
dir_output_should_be_in=video_path,
video_name=video_name,
output_suffix=output_suffix)
extract_pose(video_fullpath=video_fullpath,
output_folder=output_folder,
output_suffix=output_suffix,
view=view,
doOverwrite=doOverwrite,
progress_bar_signal=progress_bar_signal,
mars_opts=mars_opts,
verbose=verbose,
max_frames=max_frames)
return
def rename(path, ver_old):
ver = mof.get_version_suffix()
ver_old = 'v' + ver_old
for dir, subdirs, filenames in os.walk(path):
if ver_old in dir:
old_dir = dir
new_dir = dir.replace(ver_old, ver)
if not os.path.exists(new_dir): os.rename(old_dir, new_dir)
else: new_dir = dir
for numv, fname in enumerate(filenames):
if ver_old in fname:
oldf = fname
newf = fname.replace(ver_old, ver)
if os.path.exists(os.path.join(new_dir, oldf)):
os.rename(os.path.join(new_dir, oldf),
os.path.join(new_dir, newf))
def __init__(self, parent=None):
super(MainWindow, self).__init__(parent)
#widnow setup
resolution = QDesktopWidget().screenGeometry()
self.screen_w = resolution.width()
self.screen_h = resolution.height()
self.setGeometry(0, 0, 650, 200)
self.setWindowTitle('bento dumper' + mof.get_version_suffix())
self.setWindowIcon(QIcon('icons/run.png'))
#center window
qr = self.frameGeometry()
cp = QtGui.QDesktopWidget().availableGeometry().center()
qr.moveCenter(cp)
self.move(qr.topLeft())
#adjust size
self.resize(self.screen_w / 2, self.screen_h / 16)
self.Menu()
self.Layout()
central_widget = QtGui.QWidget()
central_widget.setLayout(self.main_layout)
self.setCentralWidget(central_widget)
def dump_bento(video_fullpath, output_suffix='', pose_file='', basepath=''):
if not output_suffix:
# Default suffix is just the version number.
output_suffix = mof.get_version_suffix()
video_path = os.path.dirname(video_fullpath)
video_name = os.path.basename(video_fullpath)
# Get the output folder for this specific mouse.
output_folder = mof.get_mouse_output_dir(
dir_output_should_be_in=video_path,
video_name=video_name,
output_suffix=output_suffix)
mouse_name = output_folder.split('/')[-1]
# if not movie_file:
# movie_name = mouse_name + '_Top_J85.seq'
#
# movie_location = output_folder
# movie_location = os.path.split(movie_location)[0]
# movie_location = os.path.split(movie_location)[0]
#
# movie_file = os.path.join(movie_location, movie_name)
# else:
# movie_file = os.path.basename(movie_file)
if not pose_file:
pose_basename = mof.get_pose_no_ext(video_fullpath=video_fullpath,
output_folder=output_folder,
view='top',
output_suffix=output_suffix)
top_pose_fullpath = pose_basename + '.mat'
# else:
# pose_file = os.path.basename(pose_file)
""" This function writes an xls with information for bento in it."""
wb = xlwt.Workbook(encoding='utf-8')
ws1 = wb.add_sheet('Sheet1', cell_overwrite_ok=True)
ws1.write(0, 0, basepath) # A1
ws1.write(0, 1, 'Ca framerate:') # B1
ws1.write(0, 2, 0) # C1
ws1.write(0, 3, 'Annot framerate:') # D1
ws1.write(0, 4, 30) # E1
ws1.write(0, 5, 'Multiple trials/Ca file:') # F1
ws1.write(0, 6, 0) # G1
ws1.write(0, 7, 'Multiple trails/annot file') # H1
ws1.write(0, 8, 0) # I1
ws1.write(0, 9, 'Includes behavior movies:') # J1
ws1.write(0, 10, 1) # K1
ws1.write(
0, 11,
'Offset (in seconds; positive values = annot starts before Ca):') # L1
ws1.write(0, 12, 0) # M1
ws1.write(1, 0, 'Mouse') # A2
ws1.write(1, 1, 'Sessn') # B2
ws1.write(1, 2, 'Trial') # C2
ws1.write(1, 3, 'Stim') # D2
ws1.write(1, 4, 'Calcium imaging file') # E2
ws1.write(1, 5, 'Start Ca') # F2
ws1.write(1, 6, 'Stop Ca') # G2
ws1.write(1, 7, 'FR Ca') # H2
ws1.write(1, 8, 'Alignments') # I2
ws1.write(1, 9, 'Annotation file') # J2
ws1.write(1, 10, 'Start Anno') # K2
ws1.write(1, 11, 'Stop Anno') # L2
ws1.write(1, 12, 'FR Anno') # M2
ws1.write(1, 13, 'Offset') # N2
ws1.write(1, 14, 'Behavior movie') # O2
ws1.write(1, 15, 'Tracking') # P2
ws1.write(2, 0, 1) # A2
ws1.write(2, 1, 1) # B2
ws1.write(2, 2, 1) # C2
ws1.write(2, 3, '') # D2
ws1.write(2, 4, '') # E2
ws1.write(2, 5, '') # F2
ws1.write(2, 6, '') # G2
ws1.write(2, 7, '') # H2
ws1.write(2, 8, '') # I2
ann = [
os.path.join(output_folder, f) for f in os.listdir(output_folder)
if is_gt_annotation(f) | ('pred' in f)
]
ann = sorted(ann)
ann = [get_rel_path(annot_path, basepath) for annot_path in ann]
ws1.write(2, 9, ';'.join(ann)) # J2
ws1.write(2, 10, '') # K2
ws1.write(2, 11, '') # L2
ws1.write(2, 12, '') # M2
ws1.write(2, 13, '') # N2
ws1.write(2, 14, get_rel_path(video_fullpath, basepath)) # O2
ws1.write(2, 15, get_rel_path(top_pose_fullpath, basepath)) # P2
bento_name = 'bento_' + output_suffix + '.xls'
wb.save(os.path.join(output_folder, bento_name))
return 1
def dump_bento_across_dir(root_path):
''' This function makes a bento file for a specific directory.'''
wb = xlwt.Workbook(encoding='utf-8')
ws1 = wb.add_sheet('Sheet1', cell_overwrite_ok=True)
ws1.write(0, 0, os.path.abspath(root_path)) # A1
ws1.write(0, 1, 'Ca framerate:') # B1
ws1.write(0, 2, 0) # C1
ws1.write(0, 3, 'Annot framerate:') # D1
ws1.write(0, 4, 30) # E1
ws1.write(0, 5, 'Multiple trials/Ca file:') # F1
ws1.write(0, 6, 0) # G1
ws1.write(0, 7, 'Multiple trails/annot file') # H1
ws1.write(0, 8, 0) # I1
ws1.write(0, 9, 'Includes behavior movies:') # J1
ws1.write(0, 10, 1) # K1
ws1.write(
0, 11,
'Offset (in seconds; positive values = annot starts before Ca):') # L1
ws1.write(0, 12, 0) # M1
ws1.write(0, 13, 'Includes tracking data:')
ws1.write(0, 14, 0)
ws1.write(0, 15, 'Includes audio files:')
ws1.write(0, 16, 0)
ws1.write(1, 0, 'Mouse') # A2
mouse_col_num = 0
session_col_num = 1
trial_col_num = 2
ws1.write(1, 1, 'Sessn') # B2
ws1.write(1, 2, 'Trial') # C2
ws1.write(1, 3, 'Stim') # D2
ws1.write(1, 4, 'Calcium imaging file') # E2
ws1.write(1, 5, 'Start Ca') # F2
ws1.write(1, 6, 'Stop Ca') # G2
ws1.write(1, 7, 'FR Ca') # H2
ws1.write(1, 8, 'Alignments') # I2
ws1.write(1, 9, 'Annotation file') # J2
annot_file_col_num = 9
ws1.write(1, 10, 'Start Anno') # K2
ws1.write(1, 11, 'Stop Anno') # L2
ws1.write(1, 12, 'FR Anno') # M2
ws1.write(1, 13, 'Offset') # N2
ws1.write(1, 14, 'Behavior movie') # O2
behavior_movie_col_num = 14
ws1.write(1, 15, 'Tracking') # P2
tracking_file_col_num = 15
ws1.write(1, 16, 'Audio file')
audio_file_col_num = 16
ws1.write(1, 17, 'tSNE')
# ws1.write(2, 0, 1) # A2
# ws1.write(2, 1, 1) # B2
# ws1.write(2, 2, 1) # C2
# ws1.write(2, 3, '') # D2
# ws1.write(2, 4, '') # E2
# ws1.write(2, 5, '') # F2
# ws1.write(2, 6, '') # G2
# ws1.write(2, 7, '') # H2
# ws1.write(2, 8, '') # I2
mouse_number = 0
row_num = 2
# Going through everything in this directory.
audio_filenames = []
add_audio_count = 0
nonaudio_filenames = []
for path, subdirs, filenames in os.walk(root_path):
for fname in sorted(filenames):
fname = os.path.join(path, fname)
if fname.endswith('.wav'):
audio_filenames.append(fname)
else:
nonaudio_filenames.append(fname)
audio_filenames = sorted(audio_filenames)
nonaudio_filenames = sorted(nonaudio_filenames)
for fname in nonaudio_filenames:
try:
cond1 = '.seq' not in fname
cond2 = 'skipped' in path
if (cond1) | cond2:
continue
if 'Top' in fname:
front_fname, top_fname, mouse_name = mof.get_names(fname)
fullpath_to_front = os.path.join(path, front_fname)
fullpath_to_top = os.path.join(path, top_fname)
else:
# This is a seq file, but doesnt have "Top" or "Front" in it. Let's skip it.
continue
# Add their info to the bento file at the appropriate level.
video_fullpath = fullpath_to_top
output_suffix = ''
video_path = os.path.dirname(video_fullpath)
video_name = os.path.basename(video_fullpath)
# Get the output folder for this specific mouse.
output_folder = mof.get_mouse_output_dir(
dir_output_should_be_in=video_path,
video_name=video_name,
output_suffix=output_suffix)
_, _, mouse_name = mof.get_names(video_name=video_name)
pose_basename = mof.get_pose_no_ext(video_fullpath=video_fullpath,
output_folder=output_folder,
view='top',
output_suffix=output_suffix)
top_pose_fullpath = pose_basename + '.json'
same_path_ann = [
os.path.join(root_path, f) for f in os.listdir(root_path)
if is_annotation_file(f, mouse_name)
]
ann = [
os.path.join(output_folder, f)
for f in os.listdir(output_folder)
if is_annotation_file(f, mouse_name)
]
ann = sorted(ann)
ann = [get_normrel_path(f, root_path) for f in ann]
pose_cond = os.path.exists(top_pose_fullpath)
video_cond = os.path.exists(video_fullpath)
should_write = (pose_cond and video_cond)
if should_write:
old_mouse_number = mouse_number
mouse_number = get_mouse_number(video_fullpath)
mouse_cond = (old_mouse_number == mouse_number)
# TODO: Session condition
sess_cond = (True)
if mouse_cond and sess_cond:
trial_count += 1
else:
trial_count = 1
ws1.write(row_num, mouse_col_num, mouse_number) # A2
ws1.write(row_num, session_col_num, 1) # B2
ws1.write(row_num, trial_col_num, trial_count) # C2
ws1.write(row_num, annot_file_col_num, ';'.join(ann)) # J2
ws1.write(row_num, 10, '') # K2
ws1.write(row_num, 11, '') # L2
ws1.write(row_num, 12, '') # M2
ws1.write(row_num, 13, '') # N2
track_file = get_normrel_path(top_pose_fullpath, root_path)
ws1.write(row_num, behavior_movie_col_num,
get_normrel_path(fullpath_to_top, root_path)) # O2
ws1.write(row_num, tracking_file_col_num, track_file) # P2
row_num += 1
except Exception as e:
print(e)
error_msg = 'ERROR: ' + fname + ' has failed. ' + str(e)
continue
# End of try-except block
# End of particular fname
# End of the particular root_path
last_row = row_num
row_num = 2
for audio_file_count, audio_file in enumerate(audio_filenames):
# Write the files in order.
ws1.write(row_num + audio_file_count, audio_file_col_num + 2,
get_normrel_path(audio_file, root_path))
bento_name = 'bento_' + mof.get_version_suffix() + '.xls'
wb.save(os.path.join(root_path, bento_name))
return
def create_video_results_wrapper(top_video_fullpath,classifier_path,
progress_bar_signal,
view='top',
doOverwrite=0,output_suffix=''):
try:
video_fullpath = top_video_fullpath
video_path = os.path.dirname(video_fullpath)
video_name = os.path.basename(video_fullpath)
model_type = mof.get_clf_type(classifier_path=classifier_path)
# Get the output folder for this specific mouse.
output_folder = mof.get_mouse_output_dir(dir_output_should_be_in=video_path, video_name=video_name,
output_suffix=output_suffix)
# Get the pose and feature files' names.
pose_basename = mof.get_pose_no_ext(video_fullpath=top_video_fullpath,
output_folder=output_folder,
view='top',
output_suffix=output_suffix)
top_pose_fullpath = pose_basename + '.json'
# Get the name of the text file we're going to save to.
classifier_savename = mof.get_classifier_savename(video_fullpath=top_video_fullpath,
output_folder=output_folder,
view=view,
classifier_path=classifier_path,
output_suffix=output_suffix,model_type=model_type)
# print(classifier_savename)
predictions_exists=os.path.exists(classifier_savename)
top_pose_exist = os.path.exists(top_pose_fullpath)
video_savename = classifier_savename[:-3]+'mp4'
video_exists=os.path.exists(video_savename)
# print(video_savename)
if not top_pose_exist:
raise ValueError("No pose has been extracted for this video!")
if not predictions_exists:
raise ValueError("No behavior classified for this video!")
if (not video_exists) | doOverwrite:
pool = mp.Pool(1)
manager = mp.Manager()
queue_for_progress_bar = manager.Queue(20)
result = pool.apply_async(create_mp4_prediction, (video_fullpath,top_pose_fullpath,classifier_savename,video_savename, queue_for_progress_bar))
still_frames_left = True
while still_frames_left:
progress_bar_input = queue_for_progress_bar.get()
if progress_bar_input:
still_frames_left = True
progress_bar_signal.emit(progress_bar_input[0], progress_bar_input[1])
else:
still_frames_left = False
break
pool.close()
pool.join()
result.get()
else:
print("4 - Video already exists")
return
except Exception as e:
print(e)
raise (e)
return
def extract_pose(video_fullpath,
output_folder,
output_suffix,
view,
doOverwrite,
progress_bar_signal,
mars_opts,
verbose=1,
max_frames=999999):
pose_basename = mof.get_pose_no_ext(video_fullpath=video_fullpath,
output_folder=output_folder,
view=view,
output_suffix=output_suffix)
video_name = os.path.basename(video_fullpath)
pose_mat_name = pose_basename + '.mat'
# Makes the output directory, if it doesn't exist.
mof.getdir(output_folder)
_, ext = os.path.splitext(video_fullpath)
ext = ext[1:] # get rid of the dot.
already_extracted_msg = (
'1 - Pose already extracted. Change your settings to override, if you still want to extract the pose.'
)
if not (ext in video_name):
print("File type unsupported! Aborted.")
return
try:
# coremltools on MACs doesn't interact well with multiprocessing, but access to the system GPU, even if it's not
# from NVidia, more than makes up for sequential processing. coremltools only works on MacOS Catalina and up.
major, minor, _ = get_macOS_version_info()
use_multiprocessing = not (major == 10 and minor >= 15)
if verbose:
print('1 - Extracting pose')
if (not os.path.exists(pose_mat_name)) | (doOverwrite):
if not (view == 'front') and not (view == 'top'):
raise ValueError(
'Invalid view type, please specify top or front.')
return
if verbose:
print(' creating the movie reader...')
reader = vidReader(video_fullpath)
NUM_FRAMES = reader.NUM_FRAMES
IM_H = reader.IM_H
IM_W = reader.IM_W
fps = reader.fps
medianFrame = []
if mars_opts['bgSubtract']:
if verbose:
print(' calculating background...')
medianFrame = get_median_frame(vc, 'cv2')
NUM_FRAMES = min(NUM_FRAMES, max_frames)
# unpack user-provided bounding boxes if they exist:
bboxes = [None] * NUM_FRAMES
if mars_opts['useExistingBBoxes']:
print(' Unpacking user-provided bounding boxes...')
bboxes = unpack_bbox_wrapper(mars_opts, video_fullpath, IM_W,
IM_H, NUM_FRAMES)
if verbose:
print(' Processing video for detection and pose ...')
DET_IM_SIZE = 299
POSE_IM_SIZE = 256
if use_multiprocessing:
if verbose:
print(" Creating pool...")
workers_to_use = 8
pool = mp.Pool(workers_to_use)
manager = mp.Manager()
maxsize = 5
if verbose:
print(
" Pool created with %d workers. \n Creating queues"
% workers_to_use)
# create managed queues
q_start_to_predet = manager.Queue(maxsize)
q_predet_to_det = manager.Queue(maxsize)
q_predet_to_prehm = manager.Queue(maxsize)
q_det_to_postdet = manager.Queue(maxsize)
q_postdet_to_prehm = manager.Queue(maxsize)
q_prehm_to_hm_IMG = manager.Queue(maxsize)
q_prehm_to_posthm_BBOX = manager.Queue(maxsize)
q_hm_to_posthm_HM = manager.Queue(maxsize)
q_posthm_to_end = manager.Queue(maxsize)
if verbose:
print(" Queues created. \n Linking pools")
try:
results_predet = pool.apply_async(
pre_det, (q_start_to_predet, q_predet_to_det,
q_predet_to_prehm, medianFrame, IM_H, IM_W))
results_det = pool.apply_async(
run_det,
(q_predet_to_det, q_det_to_postdet, view, mars_opts))
results_postdet = pool.apply_async(
post_det, (q_det_to_postdet, q_postdet_to_prehm))
results_prehm = pool.apply_async(
pre_hm, (q_postdet_to_prehm, q_predet_to_prehm,
q_prehm_to_hm_IMG, q_prehm_to_posthm_BBOX,
IM_W, IM_H))
results_hm = pool.apply_async(
run_hm, (q_prehm_to_hm_IMG, q_hm_to_posthm_HM, view,
mars_opts))
results_posthm = pool.apply_async(
post_hm,
(q_hm_to_posthm_HM, q_prehm_to_posthm_BBOX, IM_W, IM_H,
POSE_IM_SIZE, NUM_FRAMES, pose_basename))
except Exception as e:
print("Error starting Pools:")
print(e)
raise (e)
if verbose:
print(' Pools linked.\n Feeding data...')
if progress_bar_signal:
# Update the progress bar with the number of total frames it will be processing.
progress_bar_signal.emit(0, NUM_FRAMES)
for f in range(NUM_FRAMES):
img = reader.getFrame(f)
q_start_to_predet.put([img, bboxes[f]])
# Push through the poison pill.
q_start_to_predet.put(get_poison_pill())
if verbose:
print(" Pools Started...")
pool.close()
pool.join()
if verbose:
print(" Pools Finished. \n Saving...")
top_pose_frames = results_posthm.get()
else: # don't use multiprocessing, but process frames in batches
time_steps = True
if progress_bar_signal:
# Update the progress bar with the number of total frames it will be processing.
progress_bar_signal.emit(0, NUM_FRAMES)
if time_steps:
process_time = [0.] * 10
process_time_start = time.perf_counter()
det_black, det_white = run_det_setup(view, mars_opts)
det_prev_ok_loc, det_prev_ok_conf = post_det_setup()
pose_model = run_hm_setup(view, mars_opts)
top_pose_frames, bar = post_hm_setup(NUM_FRAMES)
current_frame_num = 0
if time_steps:
process_time[0] += time.perf_counter() - process_time_start
BATCH_SIZE = 16
in_q = [None] * BATCH_SIZE
det_b_q = [None] * BATCH_SIZE
det_w_q = [None] * BATCH_SIZE
pose_image_q = [None] * BATCH_SIZE
# """
# """
for batch in range(
(NUM_FRAMES + BATCH_SIZE - 1) // BATCH_SIZE):
batch_start = batch * BATCH_SIZE
batch_end = min(batch_start + BATCH_SIZE, NUM_FRAMES)
if time_steps:
process_time_start = time.perf_counter()
for f in range(batch_start, batch_end):
if time_steps:
process_time_start_0 = time.perf_counter()
ix = f - batch_start
img = reader.getFrame(f)
if time_steps:
process_time_1a = time.perf_counter()
process_time[
1] += process_time_1a - process_time_start_0
in_q[ix], pose_image_q[ix] = pre_det_inner(
[img, bboxes[f]], medianFrame, IM_H, IM_W)
if time_steps:
process_time_1 = time.perf_counter()
process_time[9] += process_time_1 - process_time_1a
for ix in range(batch_end - batch_start):
det_b_q[ix] = run_det_inner(in_q[ix], det_black,
mars_opts)
if time_steps:
process_time_2 = time.perf_counter()
process_time[2] += process_time_2 - process_time_1
for ix in range(batch_end - batch_start):
det_w_q[ix] = run_det_inner(in_q[ix], det_white,
mars_opts)
if time_steps:
process_time_3 = time.perf_counter()
process_time[3] += process_time_3 - process_time_2
for ix in range(batch_end - batch_start):
if time_steps:
process_time_start_1 = time.perf_counter()
det_out = post_det_inner([det_b_q[ix], det_w_q[ix]],
det_prev_ok_loc,
det_prev_ok_conf)
if time_steps:
process_time_4 = time.perf_counter()
process_time[
4] += process_time_4 - process_time_start_1
prepped_images, bboxes_confs = pre_hm_inner(
det_out, pose_image_q[ix], IM_W, IM_H)
if time_steps:
process_time_5 = time.perf_counter()
process_time[5] += process_time_5 - process_time_4
predicted_heatmaps = run_hm_inner(
prepped_images, pose_model)
if time_steps:
process_time_6 = time.perf_counter()
process_time[6] += process_time_6 - process_time_5
post_hm_inner(predicted_heatmaps, bboxes_confs, IM_W,
IM_H, POSE_IM_SIZE, NUM_FRAMES,
pose_basename, top_pose_frames, bar,
current_frame_num)
if time_steps:
process_time_7 = time.perf_counter()
process_time[7] += process_time_7 - process_time_6
# Increment the frame_number.
current_frame_num += 1
if time_steps:
process_time[8] += process_time_7 - process_time_start
if progress_bar_signal:
progress_bar_signal.emit(f, 0)
if time_steps:
NS_PER_SECOND = 1000000000
print("Process Times")
print("-----------------------------")
print(f"Setup : {process_time[0]} sec\n")
print(
f"File Read : {process_time[1] / NUM_FRAMES} sec / frame"
)
print(
f"Pre Detection : {process_time[9] / NUM_FRAMES} sec / frame"
)
print(
f"Detection (black) : {process_time[2] / NUM_FRAMES} sec / frame"
)
print(
f"Detection (white) : {process_time[3] / NUM_FRAMES} sec / frame"
)
print(
f"Post Detection : {process_time[4] / NUM_FRAMES} sec / frame"
)
print(
f"Pre Heatmap : {process_time[5] / NUM_FRAMES} sec / frame"
)
print(
f"Heatmap (pose) : {process_time[6] / NUM_FRAMES} sec / frame"
)
print(
f"Post Heatmap : {process_time[7] / NUM_FRAMES} sec / frame"
)
print(
f"Total processing : {process_time[8] / NUM_FRAMES} sec / frame"
)
top_pose_frames['keypoints'] = np.array(
top_pose_frames['keypoints'])
top_pose_frames['scores'] = np.array(top_pose_frames['scores'])
top_pose_frames['bbox'] = np.array(top_pose_frames['bbox'])
top_pose_frames['bscores'] = np.array(top_pose_frames['bscores'])
sp.savemat(pose_mat_name, top_pose_frames)
if verbose:
print("Saved.\nPose Extracted")
reader.close()
return
else:
if verbose:
print(already_extracted_msg)
return
except Exception as e:
print(e)
raise (e)
return
def classify_actions_wrapper(top_video_fullpath,
front_video_fullpath,
doOverwrite,
view,
classifier_path='',
output_suffix=''):
try:
video_fullpath = top_video_fullpath
video_path = os.path.dirname(video_fullpath)
video_name = os.path.basename(video_fullpath)
model_type = mof.get_clf_type(classifier_path=classifier_path)
# Get the output folder for this specific mouse.
output_folder = mof.get_mouse_output_dir(
dir_output_should_be_in=video_path,
video_name=video_name,
output_suffix=output_suffix)
# Get the name of the features you should be loading.
front_feat_basename = mof.get_feat_no_ext(
video_fullpath=top_video_fullpath,
output_folder=output_folder,
view='front',
output_suffix=output_suffix)
top_feat_basename = mof.get_feat_no_ext(
video_fullpath=top_video_fullpath,
output_folder=output_folder,
view='top',
output_suffix=output_suffix)
# Get the name of the text file we're going to save to.
classifier_savename = mof.get_classifier_savename(
video_fullpath=top_video_fullpath,
output_folder=output_folder,
view=view,
classifier_path=classifier_path,
output_suffix=output_suffix,
model_type=model_type)
# Make their matfile names.
if 'pcf' in classifier_path:
# more hiding pcf features shenanigans, hopefully the last time?
# top_feat_basename = top_feat_basename[:-4] + 'pcf_'+top_feat_basename[-4:]
front_feat_name = front_feat_basename + '_wnd.npz'
top_feat_name = top_feat_basename + '_wnd.npz'
else:
front_feat_name = front_feat_basename + '_wnd.npz'
top_feat_name = top_feat_basename + '_wnd.npz'
# Check that features exist.
top_feats_exist = os.path.exists(top_feat_name)
front_feats_exist = os.path.exists(front_feat_name)
classifier_savename_exist = os.path.exists(classifier_savename)
# if False:
if (not classifier_savename_exist) | doOverwrite:
# If the proper features don't exist, raise an exception. Otherwise, load them.
if (view == 'top') or (view == 'toppcf'):
if top_feats_exist:
print("loading top features")
features = mcm.load_features_from_filename(
top_feat_name=top_feat_name)
else:
print(
"Top features don't exist in the proper format/location. Aborting..."
)
raise ValueError(
top_feat_name.split('/')[-1] + " doesn't exist.")
elif view == 'topfront':
if not top_feats_exist:
print(
"Top features don't exist in the proper format/location. Aborting..."
)
raise ValueError(
top_feat_name.split('/')[-1] + " doesn't exist.")
elif not front_feats_exist:
print(
"Front features don't exist in the proper format/location. Aborting..."
)
raise ValueError(
front_feat_name.split('/')[-1] + " doesn't exist.")
else: # Both featuresets exist.
print("loading top and front features")
features = mcm.load_features_from_filename(
top_feat_name=top_feat_name,
front_feat_name=front_feat_name)
else:
print("Classifier available for top or top and front view")
raise ValueError('Classifier not available for only fron view')
# Classify the actions (get the labels back).
print("predicting labels")
predicted_labels, predicted_labels_interaction = mcm.predict_labels(
features, classifier_path)
# Dump the labels into the Caltech Behavior Annotator format.
mcm.dump_labels_CBA(predicted_labels, predicted_labels_interaction,
classifier_savename)
else:
print("3 - Predictions already exist")
return
except Exception as e:
print(e)
raise (e)
return