Commit 4cefb787 authored by Julian Treiber's avatar Julian Treiber

updated tests, pom.xml

parent 46d3230b
......@@ -18,11 +18,11 @@
<!-- .. SE-Libraries .................................................. -->
<CNNArch.version>0.3.4-SNAPSHOT</CNNArch.version>
<CNNTrain.version>0.3.9-SNAPSHOT</CNNTrain.version>
<CNNArch2X.version>0.0.5-SNAPSHOT</CNNArch2X.version>
<CNNTrain.version>0.3.10-SNAPSHOT</CNNTrain.version>
<CNNArch2X.version>0.0.6-SNAPSHOT</CNNArch2X.version>
<embedded-montiarc-math-opt-generator>0.1.6</embedded-montiarc-math-opt-generator>
<EMADL2PythonWrapper.version>0.0.2-SNAPSHOT</EMADL2PythonWrapper.version>
<!-- .. Libraries .................................................. -->
<guava.version>18.0</guava.version>
<junit.version>4.12</junit.version>
......@@ -144,7 +144,7 @@
</dependency>
</dependencies>
<!-- == PROJECT BUILD SETTINGS =========================================== -->
......
......@@ -36,6 +36,7 @@ class ${tc.fileNameWithoutEnding}:
os.remove(self._model_dir_ + self._model_prefix_ + "_" + str(i) + "_newest-symbol.json")
except OSError:
pass
if os.path.isdir(self._model_dir_):
for file in os.listdir(self._model_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
......
......@@ -7,7 +7,6 @@
self.${element.name} = gluon.nn.Conv2D(channels=${element.channels?c},
kernel_size=(${tc.join(element.kernel, ",")}),
strides=(${tc.join(element.stride, ",")}),
groups=${element.groups?c},
use_bias=${element.noBias?string("False", "True")})
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
......
......@@ -8,7 +8,6 @@
kernel_size=(${tc.join(element.kernel, ",")}),
strides=(${tc.join(element.stride, ",")}),
padding=self.${element.name}padding,
groups=${element.groups?c},
use_bias=${element.noBias?string("False", "True")})
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_Alexnet import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_Alexnet:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_Alexnet:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_CifarClassifierNetwork import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_CifarClassifierNetwork:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_CifarClassifierNetwork:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_VGG16 import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_VGG16:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_VGG16:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
......@@ -86,6 +86,53 @@ class DiceLoss(gluon.loss.Loss):
diceloss = self.dice_loss(F, pred, label)
return F.mean(loss, axis=self._batch_axis, exclude=True) + diceloss
class SoftmaxCrossEntropyLossIgnoreLabel(gluon.loss.Loss):
def __init__(self, axis=-1, from_logits=False, weight=None,
batch_axis=0, ignore_label=255, **kwargs):
super(SoftmaxCrossEntropyLossIgnoreLabel, self).__init__(weight, batch_axis, **kwargs)
self._axis = axis
self._from_logits = from_logits
self._ignore_label = ignore_label
def hybrid_forward(self, F, output, label, sample_weight=None):
if not self._from_logits:
output = F.log_softmax(output, axis=self._axis)
valid_label_map = (label != self._ignore_label)
loss = -(F.pick(output, label, axis=self._axis, keepdims=True) * valid_label_map )
loss = gluon.loss._apply_weighting(F, loss, self._weight, sample_weight)
return F.sum(loss) / F.sum(valid_label_map)
@mx.metric.register
class ACCURACY_IGNORE_LABEL(mx.metric.EvalMetric):
"""Ignores a label when computing accuracy.
"""
def __init__(self, axis=1, metric_ignore_label=255, name='accuracy',
output_names=None, label_names=None):
super(ACCURACY_IGNORE_LABEL, self).__init__(
name, axis=axis,
output_names=output_names, label_names=label_names)
self.axis = axis
self.ignore_label = metric_ignore_label
def update(self, labels, preds):
mx.metric.check_label_shapes(labels, preds)
for label, pred_label in zip(labels, preds):
if pred_label.shape != label.shape:
pred_label = mx.nd.argmax(pred_label, axis=self.axis, keepdims=True)
label = label.astype('int32')
pred_label = pred_label.astype('int32').as_in_context(label.context)
mx.metric.check_label_shapes(label, pred_label)
correct = mx.nd.sum( (label == pred_label) * (label != self.ignore_label) ).asscalar()
total = mx.nd.sum( (label != self.ignore_label) ).asscalar()
self.sum_metric += correct
self.num_inst += total
@mx.metric.register
class BLEU(mx.metric.EvalMetric):
N = 4
......@@ -221,6 +268,7 @@ class CNNSupervisedTrainer_Alexnet:
optimizer_params=(('learning_rate', 0.001),),
load_checkpoint=True,
checkpoint_period=5,
load_pretrained=False,
log_period=50,
context='gpu',
save_attention_image=False,
......@@ -265,6 +313,8 @@ class CNNSupervisedTrainer_Alexnet:
begin_epoch = 0
if load_checkpoint:
begin_epoch = self._net_creator.load(mx_context)
elif load_pretrained:
self._net_creator.load_pretrained_weights(mx_context)
else:
if os.path.isdir(self._net_creator._model_dir_):
shutil.rmtree(self._net_creator._model_dir_)
......@@ -297,6 +347,10 @@ class CNNSupervisedTrainer_Alexnet:
elif loss == 'dice_loss':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_function = DiceLoss(axis=loss_axis, weight=loss_weight, sparse_label=sparseLabel, batch_axis=batch_axis)
elif loss == 'softmax_cross_entropy_ignore_label':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_ignore_label = loss_params['loss_ignore_label'] if 'loss_ignore_label' in loss_params else None
loss_function = SoftmaxCrossEntropyLossIgnoreLabel(axis=loss_axis, ignore_label=loss_ignore_label, weight=loss_weight, batch_axis=batch_axis)
elif loss == 'l2':
loss_function = mx.gluon.loss.L2Loss()
elif loss == 'l1':
......@@ -475,7 +529,7 @@ class CNNSupervisedTrainer_Alexnet:
test_iter.reset()
metric = mx.metric.create(eval_metric, **eval_metric_params)
for batch_i, batch in enumerate(test_iter):
if True:
if True:
labels = [batch.label[i].as_in_context(mx_context) for i in range(1)]
data_ = batch.data[0].as_in_context(mx_context)
......
......@@ -86,6 +86,53 @@ class DiceLoss(gluon.loss.Loss):
diceloss = self.dice_loss(F, pred, label)
return F.mean(loss, axis=self._batch_axis, exclude=True) + diceloss
class SoftmaxCrossEntropyLossIgnoreLabel(gluon.loss.Loss):
def __init__(self, axis=-1, from_logits=False, weight=None,
batch_axis=0, ignore_label=255, **kwargs):
super(SoftmaxCrossEntropyLossIgnoreLabel, self).__init__(weight, batch_axis, **kwargs)
self._axis = axis
self._from_logits = from_logits
self._ignore_label = ignore_label
def hybrid_forward(self, F, output, label, sample_weight=None):
if not self._from_logits:
output = F.log_softmax(output, axis=self._axis)
valid_label_map = (label != self._ignore_label)
loss = -(F.pick(output, label, axis=self._axis, keepdims=True) * valid_label_map )
loss = gluon.loss._apply_weighting(F, loss, self._weight, sample_weight)
return F.sum(loss) / F.sum(valid_label_map)
@mx.metric.register
class ACCURACY_IGNORE_LABEL(mx.metric.EvalMetric):
"""Ignores a label when computing accuracy.
"""
def __init__(self, axis=1, metric_ignore_label=255, name='accuracy',
output_names=None, label_names=None):
super(ACCURACY_IGNORE_LABEL, self).__init__(
name, axis=axis,
output_names=output_names, label_names=label_names)
self.axis = axis
self.ignore_label = metric_ignore_label
def update(self, labels, preds):
mx.metric.check_label_shapes(labels, preds)
for label, pred_label in zip(labels, preds):
if pred_label.shape != label.shape:
pred_label = mx.nd.argmax(pred_label, axis=self.axis, keepdims=True)
label = label.astype('int32')
pred_label = pred_label.astype('int32').as_in_context(label.context)
mx.metric.check_label_shapes(label, pred_label)
correct = mx.nd.sum( (label == pred_label) * (label != self.ignore_label) ).asscalar()
total = mx.nd.sum( (label != self.ignore_label) ).asscalar()
self.sum_metric += correct
self.num_inst += total
@mx.metric.register
class BLEU(mx.metric.EvalMetric):
N = 4
......@@ -221,6 +268,7 @@ class CNNSupervisedTrainer_CifarClassifierNetwork:
optimizer_params=(('learning_rate', 0.001),),
load_checkpoint=True,
checkpoint_period=5,
load_pretrained=False,
log_period=50,
context='gpu',
save_attention_image=False,
......@@ -265,6 +313,8 @@ class CNNSupervisedTrainer_CifarClassifierNetwork:
begin_epoch = 0
if load_checkpoint:
begin_epoch = self._net_creator.load(mx_context)
elif load_pretrained:
self._net_creator.load_pretrained_weights(mx_context)
else:
if os.path.isdir(self._net_creator._model_dir_):
shutil.rmtree(self._net_creator._model_dir_)
......@@ -297,6 +347,10 @@ class CNNSupervisedTrainer_CifarClassifierNetwork:
elif loss == 'dice_loss':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_function = DiceLoss(axis=loss_axis, weight=loss_weight, sparse_label=sparseLabel, batch_axis=batch_axis)
elif loss == 'softmax_cross_entropy_ignore_label':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_ignore_label = loss_params['loss_ignore_label'] if 'loss_ignore_label' in loss_params else None
loss_function = SoftmaxCrossEntropyLossIgnoreLabel(axis=loss_axis, ignore_label=loss_ignore_label, weight=loss_weight, batch_axis=batch_axis)
elif loss == 'l2':
loss_function = mx.gluon.loss.L2Loss()
elif loss == 'l1':
......@@ -475,7 +529,7 @@ class CNNSupervisedTrainer_CifarClassifierNetwork:
test_iter.reset()
metric = mx.metric.create(eval_metric, **eval_metric_params)
for batch_i, batch in enumerate(test_iter):
if True:
if True:
labels = [batch.label[i].as_in_context(mx_context) for i in range(1)]
data_ = batch.data[0].as_in_context(mx_context)
......
......@@ -86,6 +86,53 @@ class DiceLoss(gluon.loss.Loss):
diceloss = self.dice_loss(F, pred, label)
return F.mean(loss, axis=self._batch_axis, exclude=True) + diceloss
class SoftmaxCrossEntropyLossIgnoreLabel(gluon.loss.Loss):
def __init__(self, axis=-1, from_logits=False, weight=None,
batch_axis=0, ignore_label=255, **kwargs):
super(SoftmaxCrossEntropyLossIgnoreLabel, self).__init__(weight, batch_axis, **kwargs)
self._axis = axis
self._from_logits = from_logits
self._ignore_label = ignore_label
def hybrid_forward(self, F, output, label, sample_weight=None):
if not self._from_logits:
output = F.log_softmax(output, axis=self._axis)
valid_label_map = (label != self._ignore_label)
loss = -(F.pick(output, label, axis=self._axis, keepdims=True) * valid_label_map )
loss = gluon.loss._apply_weighting(F, loss, self._weight, sample_weight)
return F.sum(loss) / F.sum(valid_label_map)
@mx.metric.register
class ACCURACY_IGNORE_LABEL(mx.metric.EvalMetric):
"""Ignores a label when computing accuracy.
"""
def __init__(self, axis=1, metric_ignore_label=255, name='accuracy',
output_names=None, label_names=None):
super(ACCURACY_IGNORE_LABEL, self).__init__(
name, axis=axis,
output_names=output_names, label_names=label_names)
self.axis = axis
self.ignore_label = metric_ignore_label
def update(self, labels, preds):
mx.metric.check_label_shapes(labels, preds)
for label, pred_label in zip(labels, preds):
if pred_label.shape != label.shape:
pred_label = mx.nd.argmax(pred_label, axis=self.axis, keepdims=True)
label = label.astype('int32')
pred_label = pred_label.astype('int32').as_in_context(label.context)
mx.metric.check_label_shapes(label, pred_label)
correct = mx.nd.sum( (label == pred_label) * (label != self.ignore_label) ).asscalar()
total = mx.nd.sum( (label != self.ignore_label) ).asscalar()
self.sum_metric += correct
self.num_inst += total
@mx.metric.register
class BLEU(mx.metric.EvalMetric):
N = 4
......@@ -221,6 +268,7 @@ class CNNSupervisedTrainer_VGG16:
optimizer_params=(('learning_rate', 0.001),),
load_checkpoint=True,
checkpoint_period=5,
load_pretrained=False,
log_period=50,
context='gpu',
save_attention_image=False,
......@@ -265,6 +313,8 @@ class CNNSupervisedTrainer_VGG16:
begin_epoch = 0
if load_checkpoint:
begin_epoch = self._net_creator.load(mx_context)
elif load_pretrained:
self._net_creator.load_pretrained_weights(mx_context)
else:
if os.path.isdir(self._net_creator._model_dir_):
shutil.rmtree(self._net_creator._model_dir_)
......@@ -297,6 +347,10 @@ class CNNSupervisedTrainer_VGG16:
elif loss == 'dice_loss':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_function = DiceLoss(axis=loss_axis, weight=loss_weight, sparse_label=sparseLabel, batch_axis=batch_axis)
elif loss == 'softmax_cross_entropy_ignore_label':
loss_weight = loss_params['loss_weight'] if 'loss_weight' in loss_params else None
loss_ignore_label = loss_params['loss_ignore_label'] if 'loss_ignore_label' in loss_params else None
loss_function = SoftmaxCrossEntropyLossIgnoreLabel(axis=loss_axis, ignore_label=loss_ignore_label, weight=loss_weight, batch_axis=batch_axis)
elif loss == 'l2':
loss_function = mx.gluon.loss.L2Loss()
elif loss == 'l1':
......@@ -475,7 +529,7 @@ class CNNSupervisedTrainer_VGG16:
test_iter.reset()
metric = mx.metric.create(eval_metric, **eval_metric_params)
for batch_i, batch in enumerate(test_iter):
if True:
if True:
labels = [batch.label[i].as_in_context(mx_context) for i in range(1)]
data_ = batch.data[0].as_in_context(mx_context)
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_CriticNetwork import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_CriticNetwork:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_CriticNetwork:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_RosCriticNetwork import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_RosCriticNetwork:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_RosCriticNetwork:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
import mxnet as mx
import logging
import os
import shutil
from CNNNet_CriticNetwork import Net_0
......@@ -11,6 +12,7 @@ class CNNCreator_CriticNetwork:
def __init__(self):
self.weight_initializer = mx.init.Normal()
self.networks = {}
self._weights_dir_ = None
def load(self, context):
earliestLastEpoch = None
......@@ -47,6 +49,29 @@ class CNNCreator_CriticNetwork:
return earliestLastEpoch
def load_pretrained_weights(self, context):
if os.path.isdir(self._model_dir_):
shutil.rmtree(self._model_dir_)
if self._weights_dir_ is not None:
for i, network in self.networks.items():
# param_file = self._model_prefix_ + "_" + str(i) + "_newest-0000.params"
param_file = None
if os.path.isdir(self._weights_dir_):
lastEpoch = 0
for file in os.listdir(self._weights_dir_):
if ".params" in file and self._model_prefix_ + "_" + str(i) in file:
epochStr = file.replace(".params","").replace(self._model_prefix_ + "_" + str(i) + "-","")
epoch = int(epochStr)
if epoch > lastEpoch:
lastEpoch = epoch
param_file = file
logging.info("Loading pretrained weights: " + self._weights_dir_ + param_file)
network.load_parameters(self._weights_dir_ + param_file, allow_missing=True, ignore_extra=True)
else:
logging.info("No pretrained weights available at: " + self._weights_dir_ + param_file)
def construct(self, context, data_mean=None, data_std=None):
self.networks[0] = Net_0(data_mean=data_mean, data_std=data_std)
self.networks[0].collect_params().initialize(self.weight_initializer, ctx=context)
......
VGG16 weights/VGG16
\ No newline at end of file
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment