def main():
params = Params()
process(params)
src_path = params.src_path
if not src_path:
try:
from Tkinter import Tk
except ImportError:
from tkinter import Tk
try:
src_path = Tk().clipboard_get()
except BaseException as e:
print('Tk().clipboard_get() failed: {}'.format(e))
return
src_path = src_path.replace(os.sep, '/').replace('"', '')
assert os.path.exists(src_path), "src_path does not exist: {}".format(
src_path)
src_dir = os.path.dirname(src_path)
src_name = os.path.basename(src_path)
src_name_noext, src_ext = os.path.splitext(src_name)
out_name = src_name_noext + '_' + src_ext
out_path = os.path.join(src_dir, out_name)
assert not os.path.exists(out_path), "out_path already exists"
img = cv2.imread(src_path)
img_h, img_w = img.shape[:2]
out_h = out_w = 0
if img_h > img_w:
out_h = params.out_size
else:
out_w = params.out_size
resized_img = resizeAR(img, width=out_w, height=out_h)
cv2.imwrite(out_path, resized_img)
def write(self, frame, frame_id=None, prefix=''):
if self._height or self._width:
frame = resizeAR(frame, height=self._height, width=self._width)
if frame_id is None:
self.frame_id += 1
else:
self.frame_id = frame_id
if prefix:
self.filename = '{:s}.{:s}'.format(prefix, self._ext)
else:
self.filename = self._fmt % self.frame_id + '.{}'.format(self._ext)
self.curr_file_path = os.path.join(self._save_dir, self.filename)
cv2.imwrite(self.curr_file_path, frame,
(cv2.IMWRITE_JPEG_QUALITY, 100))
def main():
params = {
'src_path': '.',
'save_path': '',
'save_root_dir': '',
'img_ext': 'jpg',
'show_img': 1,
'del_src': 0,
'start_id': 0,
'n_frames': 0,
'width': 0,
'height': 0,
'fps': 30,
# 'codec': 'FFV1',
# 'ext': 'avi',
'codec': 'H264',
'ext': 'mkv',
'out_postfix': '',
'reverse': 0,
'move_src': 0,
'use_skv': 0,
'disable_suffix': 0,
'read_in_batch': 1,
'placement_type': 1,
'recursive': 0,
}
paramparse.process_dict(params)
_src_path = params['src_path']
save_path = params['save_path']
img_ext = params['img_ext']
show_img = params['show_img']
del_src = params['del_src']
start_id = params['start_id']
n_frames = params['n_frames']
__width = params['width']
__height = params['height']
fps = params['fps']
use_skv = params['use_skv']
codec = params['codec']
ext = params['ext']
out_postfix = params['out_postfix']
reverse = params['reverse']
save_root_dir = params['save_root_dir']
move_src = params['move_src']
disable_suffix = params['disable_suffix']
read_in_batch = params['read_in_batch']
placement_type = params['placement_type']
recursive = params['recursive']
img_exts = ['.jpg', '.jpeg', '.png', '.bmp', '.tif']
src_root_dir = ''
if os.path.isdir(_src_path):
if recursive:
src_paths = [_src_path]
else:
src_files = [
k for k in os.listdir(_src_path) for _ext in img_exts
if k.endswith(_ext)
]
if not src_files:
# src_paths = [os.path.join(_src_path, k) for k in os.listdir(_src_path) if
# os.path.isdir(os.path.join(_src_path, k))]
src_paths_gen = [[
os.path.join(dirpath, d) for d in dirnames if any([
os.path.splitext(f.lower())[1] in img_exts
for f in os.listdir(os.path.join(dirpath, d))
])
] for (dirpath, dirnames,
filenames) in os.walk(_src_path, followlinks=True)]
src_paths = [
item for sublist in src_paths_gen for item in sublist
]
src_root_dir = os.path.abspath(_src_path)
else:
src_paths = [_src_path]
print('Found {} image sequence(s):\n{}'.format(len(src_paths),
pformat(src_paths)))
elif os.path.isfile(_src_path):
print('Reading source image sequences from: {}'.format(_src_path))
src_paths = [
x.strip() for x in open(_src_path).readlines() if x.strip()
]
n_seq = len(src_paths)
if n_seq <= 0:
raise SystemError(
'No input sequences found in {}'.format(_src_path))
print('n_seq: {}'.format(n_seq))
else:
raise IOError('Invalid src_path: {}'.format(_src_path))
if recursive:
print('searching for images recursively')
if reverse == 1:
print('Writing the reverse sequence')
elif reverse == 2:
print('Appending the reverse sequence')
print('placement_type: {}'.format(placement_type))
exit_prog = 0
n_src_paths = len(src_paths)
cwd = os.getcwd()
for src_id, src_path in enumerate(src_paths):
seq_name = os.path.basename(src_path)
print('\n{}/{} Reading source images from: {}'.format(
src_id + 1, n_src_paths, src_path))
src_path = os.path.abspath(src_path)
if move_src:
rel_src_path = os.path.relpath(src_path, os.getcwd())
dst_path = os.path.join(cwd, 'i2v', rel_src_path)
else:
dst_path = ''
if recursive:
src_file_gen = [[
os.path.join(dirpath, f) for f in filenames
if os.path.splitext(f.lower())[1] in img_exts
] for (dirpath, dirnames,
filenames) in os.walk(src_path, followlinks=True)]
src_files = [item for sublist in src_file_gen for item in sublist]
else:
src_files = [
k for k in os.listdir(src_path) for _ext in img_exts
if k.endswith(_ext)
]
n_src_files = len(src_files)
if n_src_files <= 0:
raise SystemError('No input frames found')
src_files.sort(key=sortKey)
print('n_src_files: {}'.format(n_src_files))
if reverse == 1:
src_files = src_files[::-1]
elif reverse == 2:
src_files += src_files[::-1]
n_src_files *= 2
_width, _height = __width, __height
if os.path.exists(save_path):
dst_mtime = os.path.getmtime(save_path)
src_mtime = os.path.getmtime(src_path)
dst_mtime_fmt = datetime.fromtimestamp(dst_mtime).strftime(
'%Y-%m-%d %H:%M:%S')
src_mtime_fmt = datetime.fromtimestamp(src_mtime).strftime(
'%Y-%m-%d %H:%M:%S')
print('Output video file already exists: {}'.format(save_path))
if dst_mtime >= src_mtime:
print(
'Last modified time: {} is not older than the source: {} so skipping it'
.format(dst_mtime_fmt, src_mtime_fmt))
save_path = ''
continue
else:
print(
'Last modified time: {} is older than the source: {} so overwriting it'
.format(dst_mtime_fmt, src_mtime_fmt))
save_dir = os.path.dirname(save_path)
if save_dir and not os.path.isdir(save_dir):
os.makedirs(save_dir)
src_images = []
print('orig: {} x {}'.format(_width, _height))
if read_in_batch:
print('reading all images in batch')
src_images = [read_image(src_path, k) for k in tqdm(src_files)]
if _height <= 0 and _width <= 0:
heights, widths = zip(*[k.shape[:2] for k in src_images])
_height, _width = max(heights), max(widths)
elif _height <= 0:
_height, _width = sizeAR(src_images[0], width=_width)
elif _width <= 0:
_height, _width = sizeAR(src_images[0], height=_height)
else:
if _height <= 0 or _width <= 0:
temp_img = cv2.imread(os.path.join(src_path, src_files[0]))
_height, _width, _ = temp_img.shape
print('inferred: {} x {}'.format(_width, _height))
if not save_path:
save_fname = os.path.basename(src_path)
if not disable_suffix:
save_fname = '{}_{}'.format(save_fname, fps)
if _height > 0 and _width > 0:
save_fname = '{}_{}x{}'.format(save_fname, _width, _height)
if out_postfix:
save_fname = '{}_{}'.format(save_fname, out_postfix)
if reverse:
save_fname = '{}_r{}'.format(save_fname, reverse)
save_path = os.path.join(os.path.dirname(src_path),
'{}.{}'.format(save_fname, ext))
if src_root_dir and save_root_dir:
save_path = save_path.replace(src_root_dir, save_root_dir)
print('save_path: {}'.format(save_path))
print('src_root_dir: {}'.format(src_root_dir))
print('save_root_dir: {}'.format(save_root_dir))
print('save_path: {}'.format(save_path))
# sys.exit()
if use_skv:
video_out = skvideo.io.FFmpegWriter(
save_path,
outputdict={
'-vcodec': 'libx264', # use the h.264 codec
'-crf':
'0', # set the constant rate factor to 0, which is lossless
'-preset':
'veryslow' # the slower the better compression, in princple, try
# other options see https://trac.ffmpeg.org/wiki/Encode/H.264
})
elif codec == 'H265':
video_out = VideoWriterGPU(save_path, fps, (_width, _height))
else:
fourcc = cv2.VideoWriter_fourcc(*codec)
video_out = cv2.VideoWriter(save_path, fourcc, fps,
(_width, _height))
if video_out is None:
raise IOError(
'Output video file could not be opened: {}'.format(save_path))
print('Saving {}x{} output video to {}'.format(_width, _height,
save_path))
frame_id = start_id
pause_after_frame = 0
print_diff = max(1, int(n_src_files / 100))
start_t = time.time()
while True:
if read_in_batch:
image = src_images[frame_id]
else:
filename = src_files[frame_id]
file_path = os.path.join(src_path, filename)
if not os.path.exists(file_path):
raise SystemError(
'Image file {} does not exist'.format(file_path))
image = cv2.imread(file_path)
image = resizeAR(image,
_width,
_height,
placement_type=placement_type)
if show_img:
cv2.imshow(seq_name, image)
k = cv2.waitKey(1 - pause_after_frame) & 0xFF
if k == 27:
exit_prog = 1
break
elif k == ord('q'):
break
elif k == 32:
pause_after_frame = 1 - pause_after_frame
if use_skv:
video_out.writeFrame(
image[:, :, ::-1]) # write the frame as RGB not BGR
else:
video_out.write(image)
frame_id += 1
if frame_id % print_diff == 0:
end_t = time.time()
try:
proc_fps = float(print_diff) / (end_t - start_t)
except:
proc_fps = 0
sys.stdout.write(
'\rDone {:d}/{:d} frames at {:.4f} fps'.format(
frame_id - start_id, n_src_files - start_id, proc_fps))
sys.stdout.flush()
start_t = end_t
if (frame_id - start_id) >= n_frames > 0:
break
if frame_id >= n_src_files:
break
sys.stdout.write('\n')
sys.stdout.flush()
if use_skv:
video_out.close() # close the writer
else:
video_out.release()
if show_img:
cv2.destroyWindow(seq_name)
if move_src or del_src:
move_or_del_files(src_path, src_files, dst_path)
save_path = ''
if exit_prog:
break
def main():
_params = {
'root_dir': '/data',
'actor_id': 5,
'start_id': 1,
'end_id': -1,
'ignored_region_only': 0,
'speed': 0.5,
'show_img': 0,
'quality': 3,
'resize': 0,
'mode': 0,
'auto_progress': 0,
}
bkg_occl_seq = [24, 28, 47, 48, 49, 54]
paramparse.process_dict(_params)
root_dir = _params['root_dir']
actor_id = _params['actor_id']
start_id = _params['start_id']
ignored_region_only = _params['ignored_region_only']
end_id = _params['end_id']
show_img = _params['show_img']
params = ParamDict().__dict__
actors = params['mot_actors']
sequences = params['mot_sequences']
actor = actors[actor_id]
actor_sequences = sequences[actor]
if end_id <= start_id:
end_id = len(actor_sequences) - 1
print('root_dir: {}'.format(root_dir))
print('actor_id: {}'.format(actor_id))
print('start_id: {}'.format(start_id))
print('end_id: {}'.format(end_id))
print('actor: {}'.format(actor))
print('actor_sequences: {}'.format(actor_sequences))
n_frames_list = []
_pause = 1
__pause = 1
for seq_id in range(start_id, end_id + 1):
seq_name = actor_sequences[seq_id]
fname = '{:s}/{:s}/Annotations/xml/{:s}.xml'.format(
root_dir, actor, seq_name)
tree = ET.parse(fname)
root = tree.getroot()
out_seq_name = 'detrac_{}_{}'.format(seq_id + 1, seq_name)
out_fname = '{:s}/{:s}/Annotations/{:s}.txt'.format(
root_dir, actor, out_seq_name)
out_fid = open(out_fname, 'w')
ignored_region_obj = tree.find('ignored_region')
n_ignored_regions = 0
for bndbox in ignored_region_obj.iter('box'):
if bndbox is None:
continue
xmin = float(bndbox.attrib['left'])
ymin = float(bndbox.attrib['top'])
width = float(bndbox.attrib['width'])
height = float(bndbox.attrib['height'])
out_fid.write('-1,-1,{:f},{:f},{:f},{:f},-1,-1,-1,-1,-1\n'.format(
xmin, ymin, width, height))
n_ignored_regions += 1
if ignored_region_only:
out_fid.close()
continue
img_dir = '{:s}/{:s}/Images/{:s}'.format(root_dir, actor, out_seq_name)
frames = glob.glob('{:s}/*.jpg'.format(img_dir))
n_frames = len(frames)
n_frames_list.append(n_frames)
seq_occluded_dict = {}
skip_seq = 0
print('Processing sequence {:d} :: {:s} n_ignored_regions: {}'.format(
seq_id, seq_name, n_ignored_regions))
for frame_obj in tree.iter('frame'):
target_list = frame_obj.find('target_list')
frame_id = int(frame_obj.attrib['num'])
if show_img:
frame_path = os.path.join(img_dir,
'image{:06d}.jpg'.format(frame_id))
frame = cv2.imread(frame_path)
obj_frame = np.copy(frame)
if frame is None:
raise IOError(
'Failed to read frame: {}'.format(frame_path))
cv2.putText(frame, str(frame_id), (10, 25),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 1,
cv2.LINE_AA)
occluded = []
occluded_dict = {}
obj_dict = {}
for obj in target_list.iter('target'):
bndbox = obj.find('box')
obj_id = int(obj.attrib['id'])
xmin = float(bndbox.attrib['left'])
ymin = float(bndbox.attrib['top'])
width = float(bndbox.attrib['width'])
height = float(bndbox.attrib['height'])
assert obj_id not in obj_dict, "duplicate object found"
obj_dict[obj_id] = (xmin, ymin, width, height)
for occ_idx, occ_obj in enumerate(obj.iter('occlusion')):
occlusion = occ_obj.find('region_overlap')
occ_status = int(occlusion.attrib['occlusion_status'])
occ_id = int(occlusion.attrib['occlusion_id'])
occ_xmin = float(occlusion.attrib['left'])
occ_ymin = float(occlusion.attrib['top'])
occ_width = float(occlusion.attrib['width'])
occ_height = float(occlusion.attrib['height'])
if occ_status == 0:
"""occluded by another obj"""
_obj_id = obj_id
_occ_id = occ_id
elif occ_status == 1:
"""occluding another obj"""
_obj_id = occ_id
_occ_id = obj_id
elif occ_status == -1:
"""occluded by background"""
""""seems extremely unreliable so ignoring"""
# _obj_id = obj_id
# _occ_id = occ_id
continue
else:
raise AssertionError(
'Invalid occ_status: {}'.format(occ_status))
# assert _obj_id not in occluded_dict, "duplicate occlusion found"
if _obj_id not in occluded_dict:
occluded_dict[_obj_id] = []
occluded_dict[_obj_id].append(
(_occ_id, occ_status, occ_xmin, occ_ymin, occ_width,
occ_height))
occluded.append((obj_id, occ_status, occ_id, occ_xmin,
occ_ymin, occ_width, occ_height))
if occ_idx > 0:
raise AssertionError(
'Multiple occluding objects found')
for obj_id in obj_dict:
xmin, ymin, width, height = obj_dict[obj_id]
xmax, ymax = xmin + width, ymin + height
obj_img = np.zeros((int(height), int(width), 1),
dtype=np.uint8)
obj_img.fill(64)
if obj_id in occluded_dict:
if show_img:
_obj_frame = np.copy(obj_frame)
drawBox(_obj_frame,
xmin,
ymin,
xmax,
ymax,
label=str(obj_id),
box_color=(255, 255, 255))
# __pause = imshow('_obj_frame', _obj_frame, __pause)
for _occluded in occluded_dict[obj_id]:
occ_id, occ_status, occ_xmin, occ_ymin, occ_width, occ_height = _occluded
occ_xmax, occ_ymax = occ_xmin + occ_width, occ_ymin + occ_height
start_x, end_x = int(occ_xmin - xmin), int(occ_xmax -
xmin)
start_y, end_y = int(occ_ymin - ymin), int(occ_ymax -
ymin)
# assert start_x >= 0 and start_y >= 0, "Invalid occlusion region start: {}".format(_occluded)
# assert end_x <= width and end_y <= height, \
# "Invalid occlusion region end: {} for obj: {}\n{}, {}".format(
# _occluded, obj_dict[obj_id], (end_x, end_y), (width, height))
start_x, start_y = max(start_x, 0), max(start_y, 0)
end_x, end_y = min(end_x, width), min(end_y, height)
obj_img[int(start_y):int(end_y),
int(start_x):int(end_x)] = 192
n_occluded_pix = np.count_nonzero(obj_img == 192)
occluded_ratio = float(n_occluded_pix) / float(
obj_img.size)
if show_img:
_obj_img = resizeAR(obj_img, 500)
_obj_frame = resizeAR(_obj_frame, 1920, 1080)
cv2.putText(
_obj_img,
str('{}-{} : {:.2f}'.format(
obj_id, occ_id, occluded_ratio)), (10, 25),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255), 1,
cv2.LINE_AA)
__pause, k = imshow(('obj_img', '_obj_frame'),
(_obj_img, _obj_frame), __pause)
if k == ord('q'):
print("skip_seq")
skip_seq = 1
break
else:
occ_id = occ_xmin = occ_ymin = occ_width = occ_height = -1
occluded_ratio = 0
out_str = '{:d},{:d},{:f},{:f},{:f},{:f},1,-1,-1,-1,{:f}\n'.format(
frame_id, obj_id, xmin, ymin, width, height,
occluded_ratio)
out_fid.write(out_str)
seq_occluded_dict[frame_id] = occluded_dict
if show_img:
drawBox(frame,
xmin,
ymin,
xmax,
ymax,
label=str(obj_id),
box_color=(255, 255, 255),
is_dotted=(occluded_ratio != 0))
if skip_seq:
break
if show_img:
for _occ in occluded:
obj_id, occ_status, occ_id, occ_xmin, occ_ymin, occ_width, occ_height = _occ
if occ_status == 1:
box_color = (0, 0, 255)
elif occ_status == 0:
box_color = (255, 0, 0)
elif occ_status == -1:
box_color = (0, 255, 0)
occ_xmax, occ_ymax = occ_xmin + occ_width, occ_ymin + occ_height
drawBox(frame,
occ_xmin,
occ_ymin,
occ_xmax,
occ_ymax,
box_color=box_color,
thickness=1,
label='{}-{}'.format(str(obj_id), str(occ_id)))
frame = resizeAR(frame, 1920, 1080)
_pause, k = imshow('frame', frame, _pause)
if k == ord('q'):
print("skip_seq")
skip_seq = 1
break
if frame_id % 100 == 0:
print('\t Done {:d}/{:d} frames'.format(frame_id, n_frames))
if skip_seq:
continue
meta_file_path = out_fname.replace('.txt', '.meta')
with open(meta_file_path, 'w') as meta_fid:
pprint(seq_occluded_dict, meta_fid)
out_fid.close()
print(n_frames_list)
# wp_start_row = 0
# if screensize[0] >= 3840:
# wp_start_col = 1920
# else:
# wp_start_col = 0
#
# src_img_desktop = resizeAR(src_img, wp_width, wp_height)
# wp_end_col = wp_start_col + src_img_desktop.shape[1]
# wp_end_row = wp_start_row + src_img_desktop.shape[0]
#
#
# src_img_desktop_full = np.zeros((screensize[1], screensize[0], 3), dtype=np.uint8)
# src_img_desktop_full[wp_start_row:wp_end_row, wp_start_col:wp_end_col, :] = src_img_desktop
if wallpaper_size != screensize:
src_img = resizeAR(src_img, screensize[0], screensize[1])
wp_fname = os.path.join(wallpaper_dir,
'wallpaper_{}.jpg'.format(frame_id))
# wp_fname = os.path.join(wallpaper_dir, 'wallpaper.bmp')
cv2.imwrite(wp_fname, src_img)
# print('screensize: {}'.format(screensize))
# print('wp_fname: {}'.format(wp_fname))
win_wallpaper_func(SPI_SETDESKWALLPAPER, 0, wp_fname, 0)
sys.stdout.write('\rDone {:d} frames '.format(frame_id - start_id))
sys.stdout.flush()
interrupt_wait.wait(interval)
frame_id = start_id
pause_after_frame = 0
while True:
filename = src_file_list[frame_id]
file_path = os.path.join(src_path, filename)
if not os.path.exists(file_path):
raise SystemError('Image file {} does not exist'.format(file_path))
src_img = cv2.imread(file_path)
h, w = [int(d * resize_factor) for d in src_img.shape[:2]]
w_rmd = w % 8
# if w_rmd != 0:
# print('Resized image width {} is not a multiple of 8'.format(w))
src_img = resizeAR(src_img, w - w_rmd, h)
w -= w_rmd
src_img = src_img[:, :, 0].squeeze()
dst_h, dst_w = h, int(w / 8)
# print('src_size: {}x{}'.format(w, h))
# print('dst_size: {}x{}'.format(dst_w, dst_h))
dst_img = np.zeros((dst_h, dst_w), dtype=np.float64)
for r in range(dst_h):
curr_col = 0
for c in range(dst_w):
curr_pix_val = 0
for k in range(8):
src_pix = src_img[r, curr_col]
def main():
_params = Params()
paramparse.process(_params)
root_dir = _params.root_dir
start_id = _params.start_id
end_id = _params.end_id
write_img = _params.write_img
write_gt = _params.write_gt
save_img = _params.save_img
save_vid = _params.save_vid
codec = _params.codec
show_img = _params.show_img
vis_height = _params.vis_height
vis_width = _params.vis_width
# default_obj_size = _params.default_obj_size
ignore_missing_gt = _params.ignore_missing_gt
ignore_missing_seg = _params.ignore_missing_seg
raad_gt = _params.raad_gt
if save_img:
if not show_img:
show_img += 2
else:
if show_img > 1:
save_img = 1
params = ParamDict()
actor = 'CTC'
actor_sequences_dict = params.sequences_ctc
actor_sequences = list(actor_sequences_dict.keys())
if end_id <= start_id:
end_id = len(actor_sequences) - 1
print('root_dir: {}'.format(root_dir))
print('start_id: {}'.format(start_id))
print('end_id: {}'.format(end_id))
print('actor: {}'.format(actor))
print('actor_sequences: {}'.format(actor_sequences))
img_exts = ('.tif', )
n_frames_list = []
_pause = 1
__pause = 1
ann_cols = ('green', 'blue', 'red', 'cyan', 'magenta', 'gold', 'purple',
'peach_puff', 'azure', 'dark_slate_gray', 'navy', 'turquoise')
out_img_tif_root_path = linux_path(root_dir, actor, 'Images_TIF')
os.makedirs(out_img_tif_root_path, exist_ok=True)
out_img_jpg_root_path = linux_path(root_dir, actor, 'Images')
os.makedirs(out_img_jpg_root_path, exist_ok=True)
if save_img:
out_vis_root_path = linux_path(root_dir, actor, 'Visualizations')
os.makedirs(out_vis_root_path, exist_ok=True)
out_gt_root_path = linux_path(root_dir, actor, 'Annotations')
os.makedirs(out_gt_root_path, exist_ok=True)
n_frames_out_file = linux_path(root_dir, actor, 'n_frames.txt')
n_frames_out_fid = open(n_frames_out_file, 'w')
_exit = 0
_pause = 1
time_stamp = datetime.now().strftime("%y%m%d_%H%M%S")
log_path = linux_path(root_dir, actor, 'log_{}.log'.format(time_stamp))
tif_root_dir = linux_path(root_dir, actor, 'tif')
assert os.path.exists(tif_root_dir), "tif_root_dir does not exist"
seq_ids = _params.seq_ids
if not seq_ids:
seq_ids = list(range(start_id, end_id + 1))
n_seq = len(seq_ids)
for __id, seq_id in enumerate(seq_ids):
seq_name = actor_sequences[seq_id]
default_obj_size = actor_sequences_dict[seq_name]
seq_img_path = linux_path(tif_root_dir, seq_name)
assert os.path.exists(seq_img_path), "seq_img_path does not exist"
seq_img_src_files = [
k for k in os.listdir(seq_img_path)
if os.path.splitext(k.lower())[1] in img_exts
]
seq_img_src_files.sort()
out_gt_fid = None
n_frames = len(seq_img_src_files)
print('seq {} / {}\t{}\t{}\t{} frames'.format(__id + 1, n_seq, seq_id,
seq_name, n_frames))
n_frames_out_fid.write("{:d}: ('{:s}', {:d}),\n".format(
seq_id, seq_name, n_frames))
n_frames_list.append(n_frames)
gt_available = 0
if not raad_gt:
print('skipping GT reading')
else:
seq_gt_path = linux_path(tif_root_dir, seq_name + '_GT', 'TRA')
if not os.path.exists(seq_gt_path):
msg = "seq_gt_path does not exist"
if ignore_missing_gt:
print(msg)
else:
raise AssertionError(msg)
else:
gt_available = 1
seq_gt_tra_file = linux_path(seq_gt_path, "man_track.txt")
if os.path.exists(seq_gt_tra_file):
out_tra_file = linux_path(out_gt_root_path,
seq_name + '.tra')
print('{} --> {}'.format(seq_gt_tra_file, out_tra_file))
if not os.path.exists(out_tra_file):
shutil.copy(seq_gt_tra_file, out_tra_file)
else:
print('skipping existing {}'.format(out_tra_file))
else:
msg = "\nseq_gt_tra_file does not exist: {}".format(
seq_gt_tra_file)
if ignore_missing_gt:
print(msg)
else:
raise AssertionError(msg)
if _params.tra_only:
continue
seg_available = 0
if not raad_gt:
print('skipping segmentation reading')
else:
seq_seg_path = linux_path(tif_root_dir, seq_name + '_ST', 'SEG')
if not os.path.exists(seq_seg_path):
print("ST seq_seg_path does not exist")
seq_seg_path = linux_path(tif_root_dir, seq_name + '_GT',
'SEG')
if not os.path.exists(seq_seg_path):
msg = "GT seq_seg_path does not exist"
if ignore_missing_seg:
print(msg)
else:
raise AssertionError(msg)
else:
seg_available = 1
else:
seg_available = 1
if write_img:
out_img_tif_dir_path = linux_path(out_img_tif_root_path, seq_name)
os.makedirs(out_img_tif_dir_path, exist_ok=True)
print('copying TIF images to {}'.format(out_img_tif_dir_path))
out_img_jpg_dir_path = linux_path(out_img_jpg_root_path, seq_name)
os.makedirs(out_img_jpg_dir_path, exist_ok=True)
print('Saving jPG images to {}'.format(out_img_jpg_dir_path))
vid_out = None
if save_img:
if save_vid:
out_vis_path = linux_path(out_vis_root_path, seq_name + '.mkv')
vid_out = cv2.VideoWriter(out_vis_path,
cv2.VideoWriter_fourcc(*codec), 30,
(vis_width, vis_height))
else:
out_vis_path = linux_path(out_vis_root_path, seq_name)
os.makedirs(out_vis_path, exist_ok=True)
print('Saving visualizations to {}'.format(out_vis_path))
from collections import OrderedDict
file_id_to_gt = OrderedDict()
obj_id_to_gt_file_ids = OrderedDict()
if gt_available:
print('reading GT from {}...'.format(seq_gt_path))
seq_gt_src_files = [
k for k in os.listdir(seq_gt_path)
if os.path.splitext(k.lower())[1] in img_exts
]
seq_gt_src_files.sort()
assert len(seq_img_src_files) == len(
seq_gt_src_files
), "mismatch between the lengths of seq_img_src_files and seq_gt_src_files"
for seq_gt_src_file in tqdm(seq_gt_src_files,
disable=_params.disable_tqdm):
seq_gt_src_file_id = ''.join(k for k in seq_gt_src_file
if k.isdigit())
file_id_to_gt[seq_gt_src_file_id] = OrderedDict()
seq_gt_src_path = os.path.join(seq_gt_path, seq_gt_src_file)
seq_gt_pil = Image.open(seq_gt_src_path)
seq_gt_np = np.array(seq_gt_pil)
gt_obj_ids = list(np.unique(seq_gt_np, return_counts=False))
gt_obj_ids.remove(0)
for obj_id in gt_obj_ids:
obj_locations = np.nonzero(seq_gt_np == obj_id)
centroid_y, centroid_x = [
np.mean(k) for k in obj_locations
]
file_id_to_gt[seq_gt_src_file_id][obj_id] = [
obj_locations, centroid_y, centroid_x
]
if obj_id not in obj_id_to_gt_file_ids:
obj_id_to_gt_file_ids[obj_id] = []
obj_id_to_gt_file_ids[obj_id].append(seq_gt_src_file_id)
if write_gt:
out_gt_path = linux_path(out_gt_root_path, seq_name + '.txt')
out_gt_fid = open(out_gt_path, 'w')
file_id_to_seg = OrderedDict()
file_id_to_nearest_seg = OrderedDict()
obj_id_to_seg_file_ids = OrderedDict()
obj_id_to_seg_sizes = OrderedDict()
obj_id_to_seg_bboxes = OrderedDict()
obj_id_to_mean_seg_sizes = OrderedDict()
obj_id_to_max_seg_sizes = OrderedDict()
all_seg_sizes = []
mean_seg_sizes = None
max_seg_sizes = None
if seg_available:
print('reading segmentations from {}...'.format(seq_seg_path))
seq_seq_src_files = [
k for k in os.listdir(seq_seg_path)
if os.path.splitext(k.lower())[1] in img_exts
]
for seq_seq_src_file in tqdm(seq_seq_src_files,
disable=_params.disable_tqdm):
seq_seq_src_file_id = ''.join(k for k in seq_seq_src_file
if k.isdigit())
file_gt = file_id_to_gt[seq_seq_src_file_id]
file_id_to_seg[seq_seq_src_file_id] = OrderedDict()
seq_seq_src_path = os.path.join(seq_seg_path, seq_seq_src_file)
seq_seg_pil = Image.open(seq_seq_src_path)
seq_seg_np = np.array(seq_seg_pil)
seg_obj_ids = list(np.unique(seq_seg_np, return_counts=False))
seg_obj_ids.remove(0)
_gt_obj_ids = list(file_gt.keys())
if len(_gt_obj_ids) != len(seg_obj_ids):
print(
"\nmismatch between the number of objects in segmentation: {} and GT: {} in {}"
.format(len(seg_obj_ids), len(_gt_obj_ids),
seq_seq_src_file))
from scipy.spatial import distance_matrix
seg_centroids = []
seg_id_to_locations = {}
for seg_obj_id in seg_obj_ids:
# obj_id = gt_obj_ids[seg_obj_id - 1]
seg_obj_locations = np.nonzero(seq_seg_np == seg_obj_id)
seg_centroid_y, seg_centroid_x = [
np.mean(k) for k in seg_obj_locations
]
seg_centroids.append([seg_centroid_y, seg_centroid_x])
seg_id_to_locations[seg_obj_id] = seg_obj_locations
gt_centroids = [[k[1], k[2]] for k in file_gt.values()]
gt_centroids = np.asarray(gt_centroids)
seg_centroids = np.asarray(seg_centroids)
gt_to_seg_dists = distance_matrix(seg_centroids, gt_centroids)
# seg_min_dist_ids = np.argmin(gt_to_seg_dists, axis=1)
# gt_min_dist_ids = np.argmin(gt_to_seg_dists, axis=0)
# unique_min_dist_ids = np.unique(seg_min_dist_ids) #
#
# assert len(unique_min_dist_ids) == len(seg_min_dist_ids), \
# "duplicate matches found between segmentation and GT objects"
# seg_to_gt_obj_ids = {
# seg_obj_id: _gt_obj_ids[seg_min_dist_ids[_id]] for _id, seg_obj_id in enumerate(seg_obj_ids)
# }
from scipy.optimize import linear_sum_assignment
seg_inds, gt_inds = linear_sum_assignment(gt_to_seg_dists)
if len(seg_inds) != len(seg_obj_ids):
print(
"only {} / {} segmentation objects assigned to GT objects"
.format(len(seg_inds), len(seg_obj_ids)))
seg_to_gt_obj_ids = {
seg_obj_ids[seg_inds[i]]: _gt_obj_ids[gt_inds[i]]
for i in range(len(seg_inds))
}
# print()
for seg_obj_id in seg_obj_ids:
seg_obj_locations = seg_id_to_locations[seg_obj_id]
_gt_obj_id = seg_to_gt_obj_ids[seg_obj_id]
min_y, min_x = [np.amin(k) for k in seg_obj_locations]
max_y, max_x = [np.amax(k) for k in seg_obj_locations]
size_x, size_y = max_x - min_x, max_y - min_y
if _gt_obj_id not in obj_id_to_seg_sizes:
obj_id_to_seg_bboxes[_gt_obj_id] = []
obj_id_to_seg_sizes[_gt_obj_id] = []
obj_id_to_seg_file_ids[_gt_obj_id] = []
obj_id_to_seg_file_ids[_gt_obj_id].append(
seq_seq_src_file_id)
file_id_to_seg[seq_seq_src_file_id][_gt_obj_id] = (
seg_obj_locations, [min_x, min_y, max_x, max_y])
obj_id_to_seg_bboxes[_gt_obj_id].append(
[seq_seq_src_file, min_x, min_y, max_x, max_y])
obj_id_to_seg_sizes[_gt_obj_id].append([size_x, size_y])
all_seg_sizes.append([size_x, size_y])
obj_id_to_mean_seg_sizes = OrderedDict({
k: np.mean(v, axis=0)
for k, v in obj_id_to_seg_sizes.items()
})
obj_id_to_max_seg_sizes = OrderedDict({
k: np.amax(v, axis=0)
for k, v in obj_id_to_seg_sizes.items()
})
print('segmentations found for {} files'.format(
len(file_id_to_seg),
# '\n'.join(file_id_to_seg.keys())
))
print('segmentations include {} objects:\n{}'.format(
len(obj_id_to_seg_bboxes),
', '.join(str(k) for k in obj_id_to_seg_bboxes.keys())))
mean_seg_sizes = np.mean(all_seg_sizes, axis=0)
max_seg_sizes = np.amax(all_seg_sizes, axis=0)
for obj_id in obj_id_to_seg_file_ids:
seg_file_ids = obj_id_to_seg_file_ids[obj_id]
seg_file_ids_num = np.asarray(
list(int(k) for k in seg_file_ids))
gt_file_ids = obj_id_to_gt_file_ids[obj_id]
# gt_file_ids_num = np.asarray(list(int(k) for k in gt_file_ids))
gt_seg_file_ids_dist = {
gt_file_id: np.abs(int(gt_file_id) - seg_file_ids_num)
for gt_file_id in gt_file_ids
}
file_id_to_nearest_seg[obj_id] = {
gt_file_id: seg_file_ids[np.argmin(_dist).item()]
for gt_file_id, _dist in gt_seg_file_ids_dist.items()
}
nearest_seg_size = OrderedDict()
for frame_id in tqdm(range(n_frames), disable=_params.disable_tqdm):
seq_img_src_file = seq_img_src_files[frame_id]
# assert seq_img_src_file in seq_gt_src_file, \
# "mismatch between seq_img_src_file and seq_gt_src_file"
seq_img_src_file_id = ''.join(k for k in seq_img_src_file
if k.isdigit())
seq_img_src_path = os.path.join(seq_img_path, seq_img_src_file)
# seq_img_pil = Image.open(seq_img_src_path)
# seq_img = np.array(seq_img_pil)
seq_img = cv2.imread(seq_img_src_path, cv2.IMREAD_UNCHANGED)
# assert (seq_img == seq_img_cv).all(), "mismatch between PIL and cv2 arrays"
# seq_img_cv_unique = np.unique(seq_img_cv)
# n_seq_img_unique_cv = len(seq_img_cv_unique)
# seq_img_float = seq_img.astype(np.float32) / 65535.
max_pix, min_pix = np.amax(seq_img), np.amin(seq_img)
seq_img_float_norm = (seq_img.astype(np.float32) -
min_pix) / (max_pix - min_pix)
seq_img_uint8 = (seq_img_float_norm * 255.).astype(np.uint8)
# seq_img_uint8 = (seq_img / 256.).astype(np.uint8)
max_pix_uint8, min_pix_uint8 = np.amax(seq_img_uint8), np.amin(
seq_img_uint8)
seq_img_unique = np.unique(seq_img)
seq_img_unique_uint8 = np.unique(seq_img_uint8)
n_seq_img_unique = len(seq_img_unique)
n_seq_img_unique_uint8 = len(seq_img_unique_uint8)
if n_seq_img_unique > n_seq_img_unique_uint8:
# print('{} :: drop in number of unique values from {} ({}, {}) to {} ({}, {})'.format(
# seq_img_src_file, n_seq_img_unique, max_pix, min_pix,
# n_seq_img_unique_uint8, max_pix_uint8, min_pix_uint8))
with open(log_path, 'a') as log_fid:
log_fid.write('{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}\n'.format(
seq_name, seq_img_src_file, n_seq_img_unique,
n_seq_img_unique_uint8, max_pix, min_pix,
max_pix_uint8, min_pix_uint8))
# print()
if write_img:
# out_img_file = os.path.splitext(seq_img_src_file)[0] + '.png'
# out_img_file_path = linux_path(out_img_tif_dir_path, out_img_file)
out_img_file_tif = os.path.splitext(
seq_img_src_file)[0] + '.tif'
out_img_file_path_tif = linux_path(out_img_tif_dir_path,
out_img_file_tif)
out_img_file_uint8 = os.path.splitext(
seq_img_src_file)[0] + '.jpg'
out_img_file_path_uint8 = linux_path(out_img_jpg_dir_path,
out_img_file_uint8)
if not os.path.exists(out_img_file_path_uint8):
# cv2.imwrite(out_img_file_path, seq_img)
cv2.imwrite(out_img_file_path_uint8, seq_img_uint8)
if not os.path.exists(out_img_file_path_tif):
# print('{} --> {}'.format(seq_img_src_path, out_img_file_path_tif))
shutil.copyfile(seq_img_src_path, out_img_file_path_tif)
if show_img:
seq_img_col = seq_img_uint8.copy()
if len(seq_img_col.shape) == 2:
seq_img_col = cv2.cvtColor(seq_img_col, cv2.COLOR_GRAY2BGR)
# seq_img_col2 = seq_img_col.copy()
seq_img_col3 = seq_img_col.copy()
if not gt_available:
if save_img:
if vid_out is not None:
vid_out.write(seq_img_col)
continue
file_gt = file_id_to_gt[seq_img_src_file_id]
# seq_gt_src_file = seq_gt_src_files[frame_id]
# assert seq_gt_src_file_id == seq_img_src_file_id, \
# "Mismatch between seq_gt_src_file_id and seq_img_src_file_id"
gt_obj_ids = list(file_gt.keys())
for obj_id in gt_obj_ids:
assert obj_id != 0, "invalid object ID"
try:
nearest_seg_file_ids = file_id_to_nearest_seg[obj_id]
except KeyError:
file_seg = {}
else:
nearest_seg_file_id = nearest_seg_file_ids[
seq_img_src_file_id]
file_seg = file_id_to_seg[nearest_seg_file_id]
obj_locations, centroid_y, centroid_x = file_gt[obj_id]
if file_seg:
xmin, ymin, xmax, ymax = file_seg[obj_id][1]
size_x, size_y = xmax - xmin, ymax - ymin
nearest_seg_size[obj_id] = (size_x, size_y)
else:
try:
size_x, size_y = nearest_seg_size[obj_id]
except KeyError:
try:
size_x, size_y = obj_id_to_mean_seg_sizes[obj_id]
except KeyError:
if mean_seg_sizes is not None:
size_x, size_y = mean_seg_sizes
else:
size_x, size_y = default_obj_size, default_obj_size
ymin, xmin = centroid_y - size_y / 2.0, centroid_x - size_x / 2.0
ymax, xmax = centroid_y + size_y / 2.0, centroid_x + size_x / 2.0
width = int(xmax - xmin)
height = int(ymax - ymin)
if show_img:
col_id = (obj_id - 1) % len(ann_cols)
col = col_rgb[ann_cols[col_id]]
drawBox(seq_img_col,
xmin,
ymin,
xmax,
ymax,
label=str(obj_id),
box_color=col)
# seq_img_col2[obj_locations] = col
if file_seg:
try:
locations, bbox = file_seg[obj_id][:2]
except KeyError:
print('weird stuff going on here')
else:
seq_img_col3[locations] = col
# min_x, min_y, max_x, max_y = bbox
# drawBox(seq_img_col3, min_x, min_y, max_x, max_y, label=str(obj_id), box_color=col)
if write_gt:
out_gt_fid.write(
'{:d},{:d},{:.3f},{:.3f},{:d},{:d},1,-1,-1,-1\n'.
format(frame_id + 1, obj_id, xmin, ymin, width,
height))
# print()
skip_seq = 0
if show_img:
# images_to_stack = [seq_img_col, seq_img_col2]
images_to_stack = [
seq_img_col,
]
if file_seg:
images_to_stack.append(seq_img_col3)
__pause = _pause
else:
__pause = _pause
seq_img_vis = stackImages(images_to_stack, sep_size=5)
seq_img_vis = resizeAR(seq_img_vis,
height=vis_height,
width=vis_width)
cv2.putText(seq_img_vis,
'{}: {}'.format(seq_name,
seq_img_src_file_id), (20, 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 1)
if show_img != 2:
cv2.imshow('seq_img_vis', seq_img_vis)
k = cv2.waitKey(1 - __pause)
if k == 32:
_pause = 1 - _pause
elif k == 27:
skip_seq = 1
break
elif k == ord('q'):
break
if save_img:
if vid_out is not None:
vid_out.write(seq_img_vis)
else:
out_vis_file = os.path.splitext(
seq_img_src_file)[0] + '.jpg'
out_vis_file_path = linux_path(out_vis_path,
out_vis_file)
cv2.imwrite(out_vis_file_path, seq_img_vis)
if skip_seq or _exit:
break
if vid_out is not None:
vid_out.release()
if out_gt_fid is not None:
out_gt_fid.close()
if _exit:
break
n_frames_out_fid.close()
(dst_width, dst_height))
elif ext in image_exts:
video_out = ImageSequenceWriter(dst_path)
else:
raise IOError('Invalid ext: {}'.format(ext))
if video_out is None:
raise IOError(
'Output video file could not be opened: {}'.format(dst_path))
print('Saving {}x{} output video to {}'.format(dst_width, dst_height,
dst_path))
video_out.write(out_img)
if show_img:
out_img_disp = resizeAR(out_img, 1920, 1080)
cv2.imshow(win_name, out_img_disp)
k = cv2.waitKey(1 - pause_after_frame) & 0xFF
if k == ord('q') or k == 27:
break
elif k == 32:
pause_after_frame = 1 - pause_after_frame
sys.stdout.write('\rDone {:d}/{:d} frames '.format(frame_id - start_id,
n_frames))
sys.stdout.flush()
if frame_id - start_id >= n_frames:
break
sys.stdout.write('\n')
print('Saving {}x{} output video to {}'.format(width, height, save_path))
frame_id = start_id
pause_after_frame = 0
print_diff = max(1, int(n_src_files / 100))
start_t = time.time()
while True:
filename = src_files[frame_id]
file_path = os.path.join(src_path, filename)
if not os.path.exists(file_path):
raise SystemError('Image file {} does not exist'.format(file_path))
image = cv2.imread(file_path)
image = resizeAR(image, width, height)
if show_img:
cv2.imshow(seq_name, image)
k = cv2.waitKey(1 - pause_after_frame) & 0xFF
if k == 27:
exit_prog = 1
break
elif k == ord('q'):
break
elif k == 32:
pause_after_frame = 1 - pause_after_frame
video_out.write(image)
frame_id += 1
if video_out is None:
raise IOError('Output video file could not be opened: {}'.format(dst_path))
print('Saving {}x{} output video to {}'.format(dst_width, dst_height, dst_path))
if start_id > 0:
print('Starting from frame_id {}'.format(start_id))
frame_id = 0
pause_after_frame = 0
frames = []
if add_headers > 0:
header_path = header_files[src_id]
header_img = cv2.imread(header_path)
header_img = resizeAR(header_img, dst_width, dst_height)
for i in range(n_header_frames):
video_out.write(header_img)
while True:
ret, image = cap.read()
if not ret:
print('\nFrame {:d} could not be read'.format(frame_id + 1))
break
frame_id += 1
if frame_id <= start_id:
continue
if video_out is None:
raise IOError('Output video file could not be opened: {}'.format(dst_path))
print('Saving {}x{} output video to {}'.format(dst_width, dst_height, dst_path))
frame_id = start_id
pause_after_frame = 0
while True:
ret, image = cap.read()
if not ret:
print('\nFrame {:d} could not be read'.format(frame_id + 1))
break
image = resizeAR(image, dst_width, dst_height)
if show_img:
cv2.imshow(seq_name, image)
k = cv2.waitKey(1 - pause_after_frame) & 0xFF
if k == ord('q') or k == 27:
break
elif k == 32:
pause_after_frame = 1 - pause_after_frame
video_out.write(image)
frame_id += 1
sys.stdout.write('\rDone {:d} frames '.format(frame_id - start_id))
sys.stdout.flush()
raise IOError(
'Output video file could not be opened: {}'.format(
dst_path))
if start_id > 0:
print('Starting from frame_id {}'.format(start_id))
frame_id = 0
pause_after_frame = 0
frames = []
if add_headers > 0:
header_path = header_files[src_id]
header_img = cv2.imread(header_path)
header_img = resizeAR(header_img,
dst_width,
dst_height,
placement_type=1)
if rotate == 1:
header_img = imutils.rotate_bound(header_img, 90)
elif rotate == -1:
header_img = imutils.rotate_bound(header_img, -90)
elif rotate == 2:
header_img = imutils.rotate_bound(header_img, 180)
elif rotate == -2:
header_img = imutils.rotate_bound(header_img, -180)
for i in range(n_header_frames):
if use_skv:
video_out.writeFrame(
header_img[:, :, ::-1]) # write the frame as RGB not BGR
def main():
params = Params()
paramparse.process(params)
root_dirs = params.root_dirs
annotations = params.annotations
save_path = params.save_path
img_ext = params.img_ext
show_img = params.show_img
del_src = params.del_src
start_id = params.start_id
n_frames = params.n_frames
width = params.width
height = params.height
fps = params.fps
codec = params.codec
ext = params.ext
out_width = params.out_width
out_height = params.out_height
grid_size = params.grid_size
sep_size = params.sep_size
only_height = params.only_height
borderless = params.borderless
preserve_order = params.preserve_order
ann_fmt = params.ann_fmt
resize_factor = params.resize_factor
recursive = params.recursive
img_seq = params.img_seq
match_images = params.match_images
if match_images:
assert img_seq, "image matching is only supported in image sequence mode"
vid_exts = ['.mkv', '.mp4', '.avi', '.mjpg', '.wmv']
image_exts = ['.jpg', '.bmp', '.png', '.tif']
min_n_sources = None
min_sources = None
min_sources_id = None
sources_list = []
for root_dir_id, root_dir in enumerate(root_dirs):
root_dir = os.path.abspath(root_dir)
if img_seq:
sources = [k for k in os.listdir(root_dir) if os.path.isdir(os.path.join(root_dir, k))]
else:
sources = [k for k in os.listdir(root_dir) if os.path.splitext(k)[1] in vid_exts]
sources.sort()
n_sources = len(sources)
if min_n_sources is None or n_sources < min_n_sources:
min_n_sources = n_sources
min_sources = sources
min_sources_id = root_dir_id
sources = [os.path.join(root_dir, k) for k in sources]
sources_list.append(sources)
if match_images:
for sources_id, sources in enumerate(sources_list):
if sources_id == min_sources_id:
continue
sources = [k for k in sources if os.path.basename(k) in min_sources]
assert len(sources) == min_n_sources, "invalid sources after filtering {}".format(sources)
sources_list[sources_id] = sources
src_paths = list(zip(*sources_list))
print('sources_list:\n{}'.format(pformat(sources_list)))
print('src_paths:\n{}'.format(pformat(src_paths)))
timestamp = datetime.now().strftime("%y%m%d_%H%M%S")
_exit = 0
for _src_path in src_paths:
_annotations = annotations
_save_path = save_path
_grid_size = grid_size
n_frames = 0
src_files = _src_path
n_videos = len(src_files)
assert n_videos > 0, 'no input videos found'
if not _save_path:
seq_dir = os.path.dirname(src_files[0])
seq_name = os.path.splitext(os.path.basename(src_files[0]))[0]
dst_path = os.path.join(seq_dir, 'stacked_{}'.format(timestamp), '{}.{}'.format(seq_name, ext))
else:
out_seq_name, out_ext = os.path.splitext(os.path.basename(_save_path))
dst_path = os.path.join(os.path.dirname(_save_path), '{}_{}{}'.format(
out_seq_name, datetime.now().strftime("%y%m%d_%H%M%S"), out_ext))
save_dir = os.path.dirname(dst_path)
if save_dir and not os.path.isdir(save_dir):
os.makedirs(save_dir)
print('Stacking: {} videos:'.format(n_videos))
print('src_files:\n{}'.format(pformat(src_files)))
if _annotations:
if len(_annotations) == 1 and _annotations[0] == 1:
_annotations = []
for i in range(n_videos):
_annotations.append(seq_names[i])
else:
assert len(_annotations) == n_videos, 'Invalid annotations: {}'.format(_annotations)
for i in range(n_videos):
if _annotations[i] == '__n__':
_annotations[i] = ''
print('Adding annotations:\n{}'.format(pformat(_annotations)))
else:
_annotations = None
if not _grid_size:
_grid_size = None
else:
_grid_size = [int(x) for x in _grid_size.split('x')]
if len(_grid_size) != 2 or _grid_size[0] * _grid_size[1] != n_videos:
raise AssertionError('Invalid grid_size: {}'.format(_grid_size))
n_frames_list = []
cap_list = []
size_list = []
seq_names = []
min_n_frames = None
min_n_frames_id = 0
for src_id, src_file in enumerate(src_files):
src_file = os.path.abspath(src_file)
seq_name = os.path.splitext(os.path.basename(src_file))[0]
seq_names.append(seq_name)
if os.path.isfile(src_file):
cap = cv2.VideoCapture()
elif os.path.isdir(src_file):
cap = ImageSequenceCapture(src_file, recursive=recursive)
else:
raise IOError('Invalid src_file: {}'.format(src_file))
if not cap.open(src_file):
raise IOError('The video file ' + src_file + ' could not be opened')
cv_prop = cv2.CAP_PROP_FRAME_COUNT
h_prop = cv2.CAP_PROP_FRAME_HEIGHT
w_prop = cv2.CAP_PROP_FRAME_WIDTH
total_frames = int(cap.get(cv_prop))
_height = int(cap.get(h_prop))
_width = int(cap.get(w_prop))
cap_list.append(cap)
n_frames_list.append(total_frames)
if min_n_frames is None or total_frames < min_n_frames:
min_n_frames = total_frames
min_n_frames_id = src_id
size_list.append((_width, _height))
if match_images:
assert all(seq_name == seq_names[0] for seq_name in seq_names), "mismatch in seq_names: {}".format(seq_names)
frames_list = [os.path.basename(k) for k in cap_list[min_n_frames_id].src_files]
for src_id, cap in enumerate(cap_list):
if src_id == min_n_frames_id:
continue
cap_list[src_id].filter_files(frames_list)
n_frames_list[src_id] = min_n_frames
frame_id = start_id
pause_after_frame = 0
video_out = None
win_name = 'stacked_{}'.format(datetime.now().strftime("%y%m%d_%H%M%S"))
min_n_frames = min(n_frames_list)
max_n_frames = max(n_frames_list)
if n_frames <= 0:
n_frames = max_n_frames
else:
if max_n_frames < n_frames:
raise IOError(
'Invalid n_frames: {} for sequence list with max_n_frames: {}'.format(n_frames, max_n_frames))
if show_img == 2:
vis_only = True
print('Running in visualization only mode')
else:
vis_only = False
while True:
images = []
valid_caps = []
valid_annotations = []
for cap_id, cap in enumerate(cap_list):
ret, image = cap.read()
if not ret:
print('\nFrame {:d} could not be read'.format(frame_id + 1))
continue
images.append(image)
valid_caps.append(cap)
if _annotations:
valid_annotations.append(_annotations[cap_id])
cap_list = valid_caps
if _annotations:
_annotations = valid_annotations
# if len(images) != n_videos:
# break
frame_id += 1
if frame_id <= start_id:
break
out_img = stackImages(images, _grid_size, borderless=borderless, preserve_order=preserve_order,
annotations=_annotations, ann_fmt=ann_fmt, only_height=only_height, sep_size=sep_size)
if resize_factor != 1:
out_img = cv2.resize(out_img, (0, 0), fx=resize_factor, fy=resize_factor)
if not vis_only:
if video_out is None:
dst_height, dst_width = sizeAR(out_img, width=out_width, height=out_height)
if '.' + ext in vid_exts:
fourcc = cv2.VideoWriter_fourcc(*codec)
video_out = cv2.VideoWriter(dst_path, fourcc, fps, (dst_width, dst_height))
elif '.' + ext in image_exts:
video_out = ImageSequenceWriter(dst_path, height=dst_height, width=dst_width)
else:
raise IOError('Invalid ext: {}'.format(ext))
if video_out is None:
raise IOError('Output video file could not be opened: {}'.format(dst_path))
print('Saving {}x{} output video to {}'.format(dst_width, dst_height, dst_path))
out_img = resizeAR(out_img, width=dst_width, height=dst_height)
video_out.write(out_img)
if show_img:
# out_img_disp = out_img
out_img_disp = resizeAR(out_img, 1280)
cv2.imshow(win_name, out_img_disp)
k = cv2.waitKey(1 - pause_after_frame) & 0xFF
if k == ord('q'):
_exit = 1
break
elif k == 27:
break
elif k == 32:
pause_after_frame = 1 - pause_after_frame
sys.stdout.write('\rDone {:d}/{:d} frames '.format(frame_id - start_id, n_frames))
sys.stdout.flush()
if frame_id - start_id >= n_frames:
break
if _exit:
break
sys.stdout.write('\n')
sys.stdout.flush()
video_out.release()
if show_img:
cv2.destroyWindow(win_name)
if not os.path.isdir(out_dir):
os.makedirs(out_dir)
fourcc = cv2.VideoWriter_fourcc(*codec)
video_out = cv2.VideoWriter(out_video, fourcc, fps, (w, h))
if video_out is None:
raise IOError(
'Output video file could not be opened: {}'.format(out_video))
print('Writing video {}'.format(out_video))
while True:
ret, image = cap.read()
if not ret:
# print('frame {} could not be read from {}'.format(i, video))
break
_h, _w = image.shape[:2]
if (_h, _w) != (h, w):
print('resizing from {} to {}'.format((_h, _w), (h, w)))
image = resizeAR(image, w, h)
video_out.write(image)
cap.release()
video_dst = os.path.join(out_dir, video_fname + video_ext)
print('moving {} to {}'.format(video, video_dst))
shutil.move(video, video_dst)
video_out.release()