def main(args):
model = evaluation_util.load_model(vars(args))
chainer.serializers.load_npz(args.lift_model, model)
cap = cv.VideoCapture(args.input if args.input else 0)
hasFrame, frame = cap.read()
if not hasFrame:
exit(0)
points = OpenPose(args).predict(args, frame)
points = [vec for vec in points]
points = [np.array(vec) for vec in points]
BODY_PARTS, POSE_PAIRS = parts(args)
points = np.float32(to36M(points, BODY_PARTS))
points = np.reshape(points, [1, -1])
points = projection_gan.pose.dataset.pose_dataset.pose_dataset_base.Normalization.normalize_2d(
points)
pose = create_pose(model, points)
out_directory = "demo_out"
os.makedirs(out_directory, exist_ok=True)
split = 10
for i in range(split):
img = evaluation_util.create_projection_img(
pose,
np.pi * float(2) * i / split)
cv.imwrite(os.path.join(out_directory, "{:02d}.png".format(i)), img)
def main(args):
model = evaluation_util.load_model(vars(args))
chainer.serializers.load_npz(args.lift_model, model)
cap = cv.VideoCapture(args.input if args.input else 0)
hasFrame, frame = cap.read()
if not hasFrame:
exit(0)
points = OpenPose(args).predict(args, frame)
points = [vec for vec in points]
points = [np.array(vec) for vec in points]
BODY_PARTS, POSE_PAIRS = parts(args)
points = to36M(points, BODY_PARTS)
points = np.reshape(points, [1, -1]).astype('f')
points_norm = projection_gan.pose.dataset.pose_dataset.pose_dataset_base.Normalization.normalize_2d(
points)
pose = create_pose(model, points_norm)
out_directory = "demo_out"
os.makedirs(out_directory, exist_ok=True)
out_img = evaluation_util.create_img(points[0], frame)
cv.imwrite(os.path.join(out_directory, 'openpose_detect.jpg'), out_img)
deg = 15
for d in range(0, 360 + deg, deg):
img = evaluation_util.create_projection_img(pose, np.pi * d / 180.)
cv.imwrite(
os.path.join(out_directory, "rot_{:03d}_degree.png".format(d)),
img)
def main(args):
model = evaluation_util.load_model(vars(args))
chainer.serializers.load_npz(args.lift_model, model)
# cap = cv.VideoCapture(args.input if args.input else 0)
# 深度推定結果ディレクトリ({動画名}_json_{実行日時}_idx00)
subdir = args.base_target
frame3d_dir = "{0}/frame3d_gan".format(subdir)
os.makedirs(frame3d_dir)
#関節位置情報ファイル
posf = open(subdir +'/pos_gan.txt', 'w')
#正規化済みOpenpose位置情報ファイル
smoothedf = open(subdir +'/smoothed_gan.txt', 'w')
start_frame_index, smoothed = openpose_utils.read_openpose_json("{0}/json".format(subdir), 0)
before_pose = None
png_lib = []
for n, (frame, xy) in enumerate(smoothed.items()):
if frame % 100 == 0:
logger.info("calc idx {0}, frame {1}".format(0, frame))
logger.debug("xy")
logger.debug(xy)
points = []
for o in range(0,len(xy),2):
points.append(np.array( [xy[o], xy[o+1]] ))
logger.debug("points pre 36m")
logger.debug(points)
BODY_PARTS, POSE_PAIRS = parts(args)
# Openpose位置情報をとりあえず出力する
for poi in points:
# logger.debug(poi)
# logger.debug(poi.dtype)
# logger.debug(poi.dtype == 'object')
if poi.dtype == 'object':
# logger.debug('poi is None')
pass
else:
# logger.debug(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write("\n")
# 2d→3dに変換
points = to36M(points, BODY_PARTS)
logger.debug("points after 36m")
logger.debug(points)
points = np.reshape(points, [1, -1]).astype('f')
logger.debug("points reshape 36m")
logger.debug(points)
points_norm = projection_gan.pose.dataset.pose_dataset.pose_dataset_base.Normalization.normalize_2d(points)
logger.debug("points_norm")
logger.debug(points_norm)
poses3d = create_pose(model, points_norm)
logger.debug("poses3d")
logger.debug(poses3d)
# Plot 3d predictions
subplot_idx, exidx = 1, 1
gs1 = gridspec.GridSpec(1, 1)
gs1.update(wspace=-0.00, hspace=0.05) # set the spacing between axes.
ax = plt.subplot(gs1[subplot_idx - 1], projection='3d')
ax.view_init(18, 280)
logger.debug(np.min(poses3d))
if np.min(poses3d) < -1000 and before_pose is not None:
poses3d = before_pose
p3d = poses3d
xs = p3d[:, 0::3]
ys = p3d[:, 1::3]
zs = p3d[:, 2::3]
# 拡大する
xs *= 600
ys *= 600
zs *= 600
# 画像の出力だけYZ反転させる
p3d_copy = copy.deepcopy(p3d)
p3d_copy[:, 1::3] *= -1
p3d_copy[:, 2::3] *= -1
# 3D画像を出力する
if level[args.verbose] <= logging.INFO:
# d = 30
# img = evaluation_util.create_img_xyz(xs, ys, zs, np.pi * d / 180.)
# cv.imwrite(os.path.join(frame3d_dir, "out_{0:012d}_{0:03d}_degree.png".format(n, d)), img)
viz.show3Dpose(p3d_copy, ax, lcolor="#9b59b6", rcolor="#2ecc71", add_labels=True)
# 各フレームの単一視点からのはINFO時のみ
pngName = os.path.join(frame3d_dir, "3d_gan_{0:012d}.png".format(n))
plt.savefig(pngName)
png_lib.append(imageio.imread(pngName))
before_pose = poses3d
# 各フレームの角度別出力はデバッグ時のみ
if level[args.verbose] == logging.DEBUG:
for azim in [0, 45, 90, 135, 180, 225, 270, 315, 360]:
ax2 = plt.subplot(gs1[subplot_idx - 1], projection='3d')
ax2.view_init(18, azim)
viz.show3Dpose(p3d, ax2, lcolor="#FF0000", rcolor="#0000FF", add_labels=True)
pngName2 = os.path.join(frame3d_dir, "debug_{0:012d}_{1:03d}.png".format(n, azim))
plt.savefig(pngName2)
# 3D関節位置情報を出力する
for o in range(0,len(p3d[0]),3):
logger.debug(str(o) + " "+ str(p3d[0][o]) +" "+ str(p3d[0][o+2]) +" "+ str(p3d[0][o+1] * -1) + ", ")
posf.write(str(o) + " "+ str(p3d[0][o]) +" "+ str(p3d[0][o+2]) +" "+ str(p3d[0][o+1] * -1) + ", ")
posf.write("\n")
# # 各角度の出力はデバッグ時のみ
# if level[args.verbose] == logging.DEBUG:
# deg = 15
# for d in range(0, 360 + deg, deg):
# img = evaluation_util.create_projection_img(pose, np.pi * d / 180.)
# cv.imwrite(os.path.join(out_sub_dir, "rot_{0:03d}_degree.png".format(d)), img)
n += 1
smoothedf.close()
posf.close()
# INFO時は、アニメーションGIF生成
if level[args.verbose] <= logging.INFO:
logger.info("creating Gif {0}/movie_smoothing_gan.gif, please Wait!".format(subdir))
imageio.mimsave('{0}/movie_smoothing_gan.gif'.format(subdir), png_lib, fps=30)
parser.add_argument('gen_path', type=str)
parser.add_argument('--row', type=int, default=6)
parser.add_argument('--col', type=int, default=6)
parser.add_argument('--action', '-a', type=str, default='')
parser.add_argument('--image', action='store_true')
args = parser.parse_args()
col, row = args.col, args.row
gen_path = args.gen_path
with open(os.path.join(os.path.dirname(gen_path), 'options.json'),
'rb') as f:
opts = json.load(f)
action = args.action if args.action else opts['action']
imgs = np.zeros((350 * col, 600 * row, 3), dtype=np.uint8)
gen = evaluation_util.load_model(opts)
chainer.serializers.load_npz(gen_path, gen)
if opts['dataset'] == 'h36m':
test = H36M(action=opts['action'],
length=1,
train=False,
use_sh_detection=opts['use_sh_detection'])
elif opts['dataset'] == 'mpii':
test = MPII(train=False, use_sh_detection=opts['use_sh_detection'])
elif opts['dataset'] == 'mpi_inf':
test = MPII3DDataset(train=False)
test_iter = chainer.iterators.SerialIterator(test,
batch_size=row,
shuffle=True,
repeat=False)
def main(args):
model = evaluation_util.load_model(vars(args))
chainer.serializers.load_npz(args.lift_model, model)
# cap = cv.VideoCapture(args.input if args.input else 0)
out_directory = "demo_out"
out_vmd_directory = "vmd_out"
os.makedirs(out_vmd_directory, exist_ok=True)
#Openpose位置情報ファイル
smoothedf = open(os.path.join(out_vmd_directory, 'smoothed.txt'), 'w')
#関節位置情報ファイル
posf = open(os.path.join(out_vmd_directory, 'pos.txt'), 'w')
# 動画を1枚ずつ画像に変換する
n = 0
cap = cv.VideoCapture(args.input if args.input else 0)
while (cap.isOpened()):
# 動画から1枚キャプチャして読み込む
flag, frame = cap.read() # Capture frame-by-frame
# キャプチャが終わっていたら終了
if flag == False: # Is a frame left?
break
print("{0} ---------------------".format(n))
points = OpenPose(args).predict(args, frame)
points = [vec for vec in points]
points = [np.array(vec) for vec in points]
BODY_PARTS, POSE_PAIRS = parts(args)
print("points pre 36m")
print(points)
# Openpose位置情報をとりあえず出力する
for poi in points:
# print(poi)
# print(poi.dtype)
# print(poi.dtype == 'object')
if poi.dtype == 'object':
# print('poi is None')
pass
else:
# print(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write("\n")
# 2d→3dに変換
points = to36M(points, BODY_PARTS)
print("points after 36m")
print(points)
points = np.reshape(points, [1, -1]).astype('f')
print("points reshape 36m")
print(points)
points_norm = projection_gan.pose.dataset.pose_dataset.pose_dataset_base.Normalization.normalize_2d(
points)
print("points_norm")
print(points_norm)
pose = create_pose(model, points_norm)
print("pose")
print(pose)
# 3D関節位置情報を出力する
for o in range(0, len(pose[0]), 3):
print(
str(o) + " " + str(pose[0][o]) + " " + str(pose[0][o + 2]) +
" " + str(pose[0][o + 1] * -1) + ", ")
posf.write(
str(o) + " " + str(pose[0][o]) + " " + str(pose[0][o + 2]) +
" " + str(pose[0][o + 1] * -1) + ", ")
posf.write("\n")
out_sub_dir = os.path.join(out_directory, "{0:012d}".format(n))
os.makedirs(out_sub_dir, exist_ok=True)
out_img = evaluation_util.create_img(points[0], frame)
cv.imwrite(os.path.join(out_sub_dir, 'openpose_detect.jpg'), out_img)
deg = 15
for d in range(0, 360 + deg, deg):
img = evaluation_util.create_projection_img(pose, np.pi * d / 180.)
cv.imwrite(
os.path.join(out_sub_dir, "rot_{0:03d}_degree.png".format(d)),
img)
n += 1
smoothedf.close()
posf.close()
def main(args):
model = evaluation_util.load_model(vars(args))
chainer.serializers.load_npz(args.lift_model, model)
# cap = cv.VideoCapture(args.input if args.input else 0)
out_directory = "demo_out"
out_vmd_directory = "vmd_out"
os.makedirs(out_vmd_directory, exist_ok=True)
#Openpose位置情報ファイル
smoothedf = open(os.path.join(out_vmd_directory, 'smoothed.txt'), 'w')
#関節位置情報ファイル
posf = open(os.path.join(out_vmd_directory, 'pos.txt'), 'w')
smoothed = read_openpose_json(args.openpose_output_dir, 0)
for n, (frame, xy) in enumerate(smoothed.items()):
logger.info("calc idx {0}, frame {1}".format(0, frame))
print("xy")
print(xy)
points = []
for o in range(0, len(xy), 2):
points.append(np.array([xy[o], xy[o + 1]]))
print("points pre 36m")
print(points)
BODY_PARTS, POSE_PAIRS = parts(args)
# Openpose位置情報をとりあえず出力する
for poi in points:
# print(poi)
# print(poi.dtype)
# print(poi.dtype == 'object')
if poi.dtype == 'object':
# print('poi is None')
pass
else:
# print(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write(' ' + str(poi[0]) + ' ' + str(poi[1]))
smoothedf.write("\n")
# 2d→3dに変換
points = to36M(points, BODY_PARTS)
print("points after 36m")
print(points)
points = np.reshape(points, [1, -1]).astype('f')
print("points reshape 36m")
print(points)
points_norm = projection_gan.pose.dataset.pose_dataset.pose_dataset_base.Normalization.normalize_2d(
points)
print("points_norm")
print(points_norm)
pose = create_pose(model, points_norm)
print("pose")
print(pose)
# 3D関節位置情報を出力する
for o in range(0, len(pose[0]), 3):
print(
str(o) + " " + str(pose[0][o]) + " " + str(pose[0][o + 2]) +
" " + str(pose[0][o + 1] * -1) + ", ")
posf.write(
str(o) + " " + str(pose[0][o]) + " " + str(pose[0][o + 2]) +
" " + str(pose[0][o + 1] * -1) + ", ")
posf.write("\n")
out_sub_dir = os.path.join(out_directory, "{0:012d}".format(n))
os.makedirs(out_sub_dir, exist_ok=True)
out_img = evaluation_util.create_img(points[0], frame)
cv.imwrite(os.path.join(out_sub_dir, 'openpose_detect.jpg'), out_img)
deg = 15
for d in range(0, 360 + deg, deg):
img = evaluation_util.create_projection_img(pose, np.pi * d / 180.)
cv.imwrite(
os.path.join(out_sub_dir, "rot_{0:03d}_degree.png".format(d)),
img)
n += 1
smoothedf.close()
posf.close()