def loadRefineNetLazy(self, net):
if isinstance(net, basestring):
if os.path.exists(net):
from net.scalenet import ScaleNet, ScaleNetParams
comrefNetParams = ScaleNetParams(type=5, nChan=1, wIn=128, hIn=128, batchSize=1, resizeFactor=2,
numJoints=1, nDims=3)
self.refineNet = ScaleNet(np.random.RandomState(23455), cfgParams=comrefNetParams)
self.refineNet.load(net)
else:
raise EnvironmentError("File not found: {}".format(net))
xend = xstart + dsize[0]
ystart = int(train_data.shape[3]/2-dsize[1]/2)
yend = ystart + dsize[1]
test_data4 = test_data[:, :, ystart:yend, xstart:xend]
print train_gt3D.max(), test_gt3D.max(), train_gt3D.min(), test_gt3D.min()
print train_data.max(), test_data.max(), train_data.min(), test_data.min()
imgSizeW = train_data.shape[3]
imgSizeH = train_data.shape[2]
nChannels = train_data.shape[1]
#############################################################################
print("create network")
batchSize = 64
poseNetParams = ScaleNetParams(type=1, nChan=nChannels, wIn=imgSizeW, hIn=imgSizeH, batchSize=batchSize,
resizeFactor=2, numJoints=1, nDims=3)
poseNet = ScaleNet(rng, cfgParams=poseNetParams)
poseNetTrainerParams = ScaleNetTrainerParams()
poseNetTrainerParams.use_early_stopping = False
poseNetTrainerParams.batch_size = batchSize
poseNetTrainerParams.learning_rate = 0.0005
poseNetTrainerParams.weightreg_factor = 0.0001
poseNetTrainerParams.force_macrobatch_reload = True
poseNetTrainerParams.para_augment = True
poseNetTrainerParams.augment_fun_params = {'fun': 'augment_poses', 'args': {'normZeroOne': False,
'di': di,
'aug_modes': aug_modes,
'hd': HandDetector(train_data[0, 0].copy(), abs(di.fx), abs(di.fy), importer=di)}}
print("setup trainer")
nChan=1,
wIn=128,
hIn=128,
batchSize=1,
numJoints=14,
nDims=3)
poseNet = PoseRegNet(numpy.random.RandomState(23455),
cfgParams=poseNetParams)
poseNet.load("./NYU_network_prior.pkl")
# comrefNetParams = ScaleNetParams(type=1, nChan=1, wIn=128, hIn=128, batchSize=1, resizeFactor=2, numJoints=1, nDims=3)
# comrefNet = ScaleNet(numpy.random.RandomState(23455), cfgParams=comrefNetParams)
# comrefNet.load("./net_ICVL_COM.pkl")
comrefNetParams = ScaleNetParams(type=1,
nChan=1,
wIn=128,
hIn=128,
batchSize=1,
resizeFactor=2,
numJoints=1,
nDims=3)
comrefNet = ScaleNet(numpy.random.RandomState(23455),
cfgParams=comrefNetParams)
comrefNet.load("./net_NYU_COM.pkl")
config = {'fx': 588., 'fy': 587., 'cube': (300, 300, 300)}
# config = {'fx': 241.42, 'fy': 241.42, 'cube': (250, 250, 250)}
# config = {'fx': 224.5, 'fy': 230.5, 'cube': (300, 300, 300)} # Creative Gesture Camera
# di = ICVLImporter("./capture/")
# di.fx = 224.5
# di.fy = 230.5
# di.ux = 160.
# di.uy = 120.
rtp = RealtimeHandposePipeline(poseNet, config, di, comrefNet)
Seq2 = di.loadSequence('test_1')
testSeqs = [Seq2]
# load trained network
poseNetParams = ResNetParams(type=1,
nChan=1,
wIn=128,
hIn=128,
batchSize=1,
numJoints=14,
nDims=3)
poseNetParams.loadFile = "./eval/NYU_network_prior.pkl"
comrefNetParams = ScaleNetParams(type=1,
nChan=1,
wIn=128,
hIn=128,
batchSize=1,
resizeFactor=2,
numJoints=1,
nDims=3)
comrefNetParams.loadFile = "./eval/net_NYU_COM_AUGMENT.pkl"
config = {'fx': 588., 'fy': 587., 'cube': (300, 300, 300)}
# config = {'fx': 241.42, 'fy': 241.42, 'cube': (250, 250, 250)}
# config = {'fx': 224.5, 'fy': 230.5, 'cube': (300, 300, 300)} # Creative Gesture Camera
rtp = RealtimeHandposePipeline(poseNetParams,
config,
di,
verbose=False,
comrefNet=comrefNetParams)
# use filenames
filenames = []
di = MSRA15Importer('/content/drive/My Drive/KNOWLEDGE ENGINEERING/KE Semester 4/Core Course/CA2 (Matthew)/cvpr15_MSRAHandGestureDB/')
Seq2 = di.loadSequence('P0')
testSeqs = [Seq2]
# di = ICVLImporter('../data/ICVL/')
# Seq2 = di.loadSequence('test_seq_1')
# testSeqs = [Seq2]
#di = NYUImporter('../data/NYU/')
#Seq2 = di.loadSequence('test_1')
#testSeqs = [Seq2]
# load trained network
poseNetParams = ResNetParams(type=1, nChan=1, wIn=128, hIn=128, batchSize=1, numJoints=14, nDims=3)
poseNetParams.loadFile = "/content/deep-prior-pp/src/eval/MSRA_network_prior_0.pkl"
comrefNetParams = ScaleNetParams(type=1, nChan=1, wIn=128, hIn=128, batchSize=1, resizeFactor=2, numJoints=1, nDims=3)
comrefNetParams.loadFile = "/content/deep-prior-pp/src/eval/net_MSRA15_COM_AUGMENT.pkl"
config = {'fx': 588., 'fy': 587., 'cube': (300, 300, 300)}
# config = {'fx': 241.42, 'fy': 241.42, 'cube': (250, 250, 250)}
# config = {'fx': 224.5, 'fy': 230.5, 'cube': (300, 300, 300)} # Creative Gesture Camera
rtp = RealtimeHandposePipeline(poseNetParams, config, di, verbose=False, comrefNet=comrefNetParams)
# use filenames
filenames = []
for i in testSeqs[0].data:
filenames.append(i.fileName)
dev = FileDevice(filenames, di)
# use depth camera
# dev = CreativeCameraDevice(mirror=True)
rtp.processVideoThreaded(dev)
xend = xstart + dsize[0]
ystart = int(train_data.shape[3] / 2 - dsize[1] / 2)
yend = ystart + dsize[1]
train_data2 = train_data[:, :, ystart:yend, xstart:xend]
dsize = (int(train_data.shape[2] // 4), int(train_data.shape[3] // 4))
xstart = int(train_data.shape[2] / 2 - dsize[0] / 2)
xend = xstart + dsize[0]
ystart = int(train_data.shape[3] / 2 - dsize[1] / 2)
yend = ystart + dsize[1]
train_data4 = train_data[:, :, ystart:yend, xstart:xend]
comrefNetParams = ScaleNetParams(type=1,
nChan=1,
wIn=96,
hIn=96,
batchSize=1,
resizeFactor=2,
numJoints=1,
nDims=3)
comrefNetParams.loadFile = "../../ptm/net_MSRA15_COM_AUGMENT.pkl"
poseNet = ScaleNet(numpy.random.RandomState(23455), cfgParams=comrefNetParams)
train_data = numpy.ndarray.astype(train_data, dtype='float64')
train_data2 = numpy.ndarray.astype(train_data2, dtype='float64')
train_data4 = numpy.ndarray.astype(train_data4, dtype='float64')
#Seq_all list of sequence data
gt3D = []
for i in xrange(len(Seq_all)):
gt3D_temp = [
j.gt3Dorig[di.crop_joint_idx].reshape(1, 3) for j in Seq_all[i].data
]
gt3D.extend(gt3D_temp)
