# """Human3.6H validation samples."""
# h36m_val = BatchLoader(h36m, ['frame'],
# ['pose_w', 'pose_uvd', 'afmat', 'camera', 'action'], VALID_MODE,
# batch_size=h36m.get_length(VALID_MODE), shuffle=True)
# printcn(OKBLUE, 'Preloading Human3.6M validation samples...')
# [x_val], [pw_val, puvd_val, afmat_val, scam_val, action] = h36m_val[0]
#
# h36m_callback = H36MEvalCallback(x_val, pw_val, afmat_val,
# puvd_val[:,0,2], scam_val, action, logdir=logdir)
model = spnet.build(cfg)
loss = pose_regression_loss('l1l2bincross', 0.01)
model.compile(loss=loss, optimizer=RMSprop(lr=start_lr))
model.summary()
callbacks = []
callbacks.append(SaveModel(weights_path))
callbacks.append(mpii_callback)
# callbacks.append(h36m_callback)
steps_per_epoch = mpii.get_length(TRAIN_MODE) // batch_size_mpii
model.fit_generator(data_tr,
steps_per_epoch=steps_per_epoch,
epochs=60,
callbacks=callbacks,
workers=8,
initial_epoch=0)
loss = pose_regression_loss('l1l2bincross', 0.01)
model.compile(loss=loss, optimizer=RMSprop())
model.summary()
def lr_scheduler(epoch, lr):
if epoch in [80, 100]:
newlr = 0.2 * lr
printcn(WARNING, 'lr_scheduler: lr %g -> %g @ %d' % (lr, newlr, epoch))
else:
newlr = lr
printcn(OKBLUE, 'lr_scheduler: lr %g @ %d' % (newlr, epoch))
return newlr
callbacks = []
callbacks.append(SaveModel(weights_file))
callbacks.append(LearningRateScheduler(lr_scheduler))
callbacks.append(eval_callback)
steps_per_epoch = mpii.get_length(TRAIN_MODE) // batch_size
model.fit_generator(data_tr,
steps_per_epoch=steps_per_epoch,
epochs=120,
callbacks=callbacks,
workers=4,
initial_epoch=0)
[mpii_x_val], [mpii_p_val, mpii_afmat_val, mpii_head_val] = mpii_val[0]
"""Human3.6H validation samples."""
# h36m_val = BatchLoader(h36m, ['frame'],
# ['pose_w', 'pose_uvd', 'afmat', 'camera', 'action'], VALID_MODE,
# batch_size=h36m.get_length(VALID_MODE), shuffle=False)
# printcn(OKBLUE, 'Preloading Human3.6M validation samples...')
# [h36m_x_val], [h36m_pw_val, h36m_puvd_val, h36m_afmat_val, h36m_scam_val, \
# h36m_action] = h36m_val[0]
"""NTU subset of testing samples"""
ntu_te = BatchLoader(ntu_s1, ['frame'], ['ntuaction'],
TEST_MODE,
batch_size=1,
shuffle=False)
"""Save model callback."""
save_model = SaveModel(os.path.join(logdir,
'weights_3dp+ntu_ar_{epoch:03d}.hdf5'),
model_to_save=full_model)
def prepare_training(pose_trainable, lr):
optimizer = SGD(lr=lr, momentum=0.9, nesterov=True)
# optimizer = RMSprop(lr=lr)
models = compile_split_models(full_model,
cfg,
optimizer,
pose_trainable=pose_trainable,
ar_loss_weights=action_weight,
copy_replica=cfg.pose_replica)
full_model.summary()
"""Create validation callbacks."""
# mpii_callback = MpiiEvalCallback(mpii_x_val, mpii_p_val, mpii_afmat_val,
# """
loss = pose_regression_loss('l1l2bincross', 0.01)
model.compile(loss=loss, optimizer=RMSprop())
# model.summary()
def lr_scheduler(epoch, lr):
if epoch in [300, 350] or epoch in [50,60]:
newlr = 0.99*lr
printcn(WARNING, 'lr_scheduler: lr %g -> %g @ %d' % (lr, newlr, epoch))
else:
newlr = lr
printcn(OKBLUE, 'lr_scheduler: lr %g @ %d' % (newlr, epoch))
return newlr
callbacks = []
callbacks.append(SaveModel(weights_file,save_after_num_epoch=20))
callbacks.append(LearningRateScheduler(lr_scheduler))
steps_per_epoch = merl.get_length(TRAIN_MODE) // batch_size
model.fit_generator(data_tr,
# steps_per_epoch=steps_per_epoch,
epochs=1000,
callbacks=callbacks,
workers=8,
initial_epoch=0)
# """
[mpii_x_val], [mpii_p_val, mpii_afmat_val, mpii_head_val] = mpii_val[0]
"""Human3.6H validation samples."""
# h36m_val = BatchLoader(h36m, ['frame'],
# ['pose_w', 'pose_uvd', 'afmat', 'camera', 'action'], VALID_MODE,
# batch_size=h36m.get_length(VALID_MODE), shuffle=False)
# printcn(OKBLUE, 'Preloading Human3.6M validation samples...')
# [h36m_x_val], [h36m_pw_val, h36m_puvd_val, h36m_afmat_val, h36m_scam_val, \
# h36m_action] = h36m_val[0]
"""NTU subset of testing samples"""
ntu_te = BatchLoader(ntu_s1, ['frame'], ['ntuaction'],
TEST_MODE,
batch_size=1,
shuffle=False)
"""Save model callback."""
save_model = SaveModel(os.path.join(
"C:\\networks\\deephar\\output\\spnet\\0429\\",
'weights_3dp+ntu_ar_{epoch:03d}.hdf5'),
model_to_save=full_model)
def prepare_training(pose_trainable, lr):
optimizer = SGD(lr=lr, momentum=0.9, nesterov=True)
# optimizer = RMSprop(lr=lr)
models = compile_split_models(full_model,
cfg,
optimizer,
pose_trainable=pose_trainable,
ar_loss_weights=action_weight,
copy_replica=cfg.pose_replica)
full_model.summary()
"""Create validation callbacks."""
# mpii_callback = MpiiEvalCallback(mpii_x_val, mpii_p_val, mpii_afmat_val,
start_lr = 0.01
action_weight = 0.1
#batch_size_mpii = 3
#batch_size_h36m = 4
#batch_size_ntu = 6 #1
#batch_clips = 1 # 8/4
"""Build the full model"""
full_model = spnet.build(cfg)
"""Load pre-trained weights from pose estimation and copy replica layers."""
full_model.load_weights(
"E:\\Bachelorarbeit-SS20\\weights\\deephar\\output\\ntu_baseline\\0603\\weights_posebaseline_060.hdf5",
by_name=True)
"""Save model callback."""
save_model = SaveModel(os.path.join(
"E:\\Bachelorarbeit-SS20\\weights\\deephar\\output\\spnet\\0625",
'weights_3dp+benset_ar_{epoch:03d}.hdf5'),
model_to_save=full_model)
#with open("C:\\networks\\deephar\\output\\split_train_val_benset\\train_list.p", 'rb') as fp:
# train_data_keys = pickle.load(fp)
#with open("C:\\networks\\deephar\\output\\split_train_val_benset\\val_list.p", 'rb') as fp:
# val_data_keys = pickle.load(fp)
sys.path.append(os.path.join(os.getcwd(), 'benset'))
from benset_batchloader_benset import *
from benset_dataloader_ar import *
dataset_path_green = "E:\\Bachelorarbeit-SS20\\datasets\\Benset256_green"