def main():
Data_root = "ILSVRC2015/Data/VID/train/"
source_root = "ILSVRC2015/ImageSets/Videos/train.txt"
batch_size = 100
GPU = 1
output_path = 'output/colornet/frame_all'
imageiter = imageIter.ImageIter(batch_size=batch_size,
source_root=source_root,
Data_root=Data_root)
sym_instance = color_flow.color_flow()
sym = sym_instance.get_train_symbol()
#feat_sym = sym.get_internals()
#feat_sym = sym.get_internals('prior_boost_nongray_output')
summary(sym, shape=(imageiter.k, 3, 176, 176))
# load param
color_pretrain = './model/pretrained_model/colornet'
#color_pretrain = 'output/colornet/colornet100'
epoch = 0
arg_params, aux_params = load_model.load_param(color_pretrain, epoch)
#arg_params=None
#aux_params=None
sym_instance.init_weight(arg_params, aux_params)
# metric
color_metric = metric.ColorMetric()
eval_metrics = mx.metric.CompositeEvalMetric()
for child_metric in [color_metric]:
eval_metrics.add(child_metric)
batch_end_callback = callback.Speedometer(batch_size=1, frequent=10)
checkpoint = mx.callback.do_checkpoint('output/colornet/colornet_all', 1)
optimizer = mx.optimizer.Adam(learning_rate=3.16e-5,
beta1=0.9,
beta2=0.99,
rescale_grad=1.0 / imageiter.k)
model = mx.module.Module(context=mx.gpu(0), symbol=sym, label_names=None)
print 'training the data'
model.fit(imageiter, begin_epoch=0, num_epoch=3, \
batch_end_callback = batch_end_callback, \
epoch_end_callback = checkpoint, \
arg_params=arg_params, aux_params=aux_params, \
optimizer=optimizer, \
kvstore='local', \
eval_metric=eval_metrics)#, \
img_root = 'ILSVRC2015/Data/VID/val/ILSVRC2015_val_00' + sample + '/'
for idx in range(len(List)):
List[idx] = List[idx].strip()
sym_instance = CFNet.CFNet()
sym = sym_instance.get_flow_test_symbol()
#summary(sym, (2,1,224,224))
prefix = './output/colornet/CFNwarp100'
#prefix = 'model/pretrained_model/colornet'
#epoch = 2
epoch = 0
#param_file = '%s-0060.params' % prefix
#symbol_file = '%s' % prefix
print 'loading the params'
arg_params, aux_params = load_model.load_param(prefix,
epoch,
process=True)
sym_instance.init_weight(arg_params, aux_params)
print 'init the module'
model = mx.module.Module(context=mx.gpu(GPU),
symbol=sym,
data_names=['data', 'data_cur'],
label_names=None)
model.bind(data_shapes=[('data', (1, 1, 224, 224)),
('data_cur', (1, 1, 224, 224))],
label_shapes=None,
for_training=False)
model.init_params(arg_params=arg_params, aux_params=aux_params)
print 'test the data'
H_in = 224
W_in = 224
def main():
Data_root = "ILSVRC2015/Data/VID/train/"
source_root = "ILSVRC2015/ImageSets/Videos/train.txt"
batch_size = 90
output_path = 'output/colornet/CFN_all'
GPU = 0
imageiter = imageIter.ImageIter(batch_size=batch_size,
source_root=source_root,
Data_root=Data_root)
sym_instance = CFNet.CFNet()
sym = sym_instance.get_train_symbol()
#mx.visualization.plot_network(sym).view()
summary(sym, shape=(batch_size, 3, 176, 176))
# load param
#color_pretrain = './model/pretrained_model/colornet'
color_pretrain = 'output/colornet/CFNwarp100'
epoch = 0
arg_params, aux_params = load_model.load_param(color_pretrain, epoch)
fixed_param_names = [
'bw_conv1_1_weight', 'bw_conv1_1_bias', 'conv1_2_weight',
'conv1_2_bias', 'conv2_1_weight', 'conv2_2_bias', 'conv2_2_weight',
'conv2_2_bias', 'conv3_1_weight', 'conv3_1_bias', 'conv3_2_weight',
'conv3_2_bias', 'conv3_3_weight', 'conv3_3_bias', 'conv4_1_weight',
'conv4_1_bias', 'conv4_2_weight', 'conv4_2_bias', 'conv4_3_weight',
'conv4_3_bias', 'conv5_1_weight', 'conv5_1_bias', 'conv5_2_weight',
'conv5_2_bias', 'conv5_3_weight', 'conv5_3_bias', 'conv6_1_weight',
'conv6_1_bias', 'conv6_2_weight', 'conv6_2_bias', 'conv6_3_weight',
'conv6_3_bias', 'conv7_1_weight', 'conv7_1_bias', 'conv7_2_weight',
'conv7_2_bias', 'conv7_3_weight', 'conv7_3_bias', 'conv8_1_weight',
'conv8_2_weight', 'conv8_2_bias', 'conv8_3_weight', 'conv8_3_bias',
'conv8_313_weight', 'data_ab_ss_weight', 'data_ab_ss_bias',
'conv1_2norm_beta', 'conv1_2norm_gamma', 'conv2_2norm_beta',
'conv2_2norm_gamma', 'conv3_3norm_beta', 'conv3_3norm_gamma',
'conv4_3norm_beta', 'conv4_3norm_gamma', 'conv5_3norm_beta',
'conv5_3norm_gamma', 'conv6_3norm_beta', 'conv6_3norm_gamma',
'conv7_3norm_beta', 'conv7_3norm_gamma'
]
#fixed_param_names = ['data_ab_ss', 'data_ab_ss_cur']
sym_instance.init_weight(arg_params, aux_params)
# metric
color_metric = metric.ColorMetric()
flow_metric = metric.FlowMetric()
eval_metrics = mx.metric.CompositeEvalMetric()
for child_metric in [color_metric, flow_metric]:
eval_metrics.add(child_metric)
batch_end_callback = callback.Speedometer(batch_size=1, frequent=1)
optimizer = mx.optimizer.Adam(learning_rate=3.16e-5,
beta1=0.9,
beta2=0.99,
rescale_grad=1.0 / batch_size)
data_names = ['data']
model = mx.module.Module(context=mx.gpu(GPU),
symbol=sym,
data_names=data_names,
label_names=None)
checkpoint = mx.callback.module_checkpoint(model,
output_path,
period=1,
save_optimizer_states=False)
print 'training the data'
model.fit(imageiter, begin_epoch=0, num_epoch=10, \
batch_end_callback = batch_end_callback, \
epoch_end_callback = checkpoint, \
arg_params=arg_params, aux_params=aux_params, \
optimizer=optimizer, \
kvstore='local', \
eval_metric=eval_metrics)
#img_root = 'ILSVRC2015/Data/VID/train/ILSVRC2015_VID_train_0000/ILSVRC2015_train_00139005/'
img_root = 'ILSVRC2015/Data/VID/val/ILSVRC2015_val_00'+sample+'/'
for i in range(len(List)):
List[i] = List[i].strip()
sym_instance = color_flow.color_flow()
sym = sym_instance.get_test_symbol()
summary(sym, (1,1,224,224))
#prefix = './output/colornet/CFN100'
prefix = 'model/pretrained_model/colornet'
#epoch = 2
epoch = 0
#param_file = '%s-0060.params' % prefix
#symbol_file = '%s' % prefix
print 'loading the params'
arg_params, aux_params = load_model.load_param(prefix, epoch)
sym_instance.init_weight(arg_params, aux_params)
#sym, arg_params, aux_params = mx.model.load_checkpoint(symbol_file, param_file)
print 'init the module'
model = mx.module.Module(context=mx.gpu(GPU), symbol=sym, label_names=None)
model.bind(data_shapes=[('data',(1,1,224,224))], label_shapes=None,for_training=False)
model.init_params(arg_params=arg_params, aux_params=aux_params, allow_missing=False, force_init=True)
#model.set_params(arg_params=arg_params, aux_params=aux_params, allow_missing=False, force_init=True)
print 'test the data'
H_in = 224
W_in = 224
H_out = 56
W_out = 56
#for i in range(2):
for i in range(len(List)):
print img_root + List[i].strip()