def test_convert():
fName = '/work4/pulkitag-code/code/ief/IEF/models/new_models/models/new-model.mat'
outName = 'try.prototxt'
model = MatConvNetModel(fName)
imgLayer = mpu.get_layerdef_for_proto('DeployData', 'image', None,
**{'ipDims': [1, 3, 224, 224]})
kpLayer = mpu.get_layerdef_for_proto('DeployData', 'kp_pos', None,
**{'ipDims': [1, 17, 2, 1]})
lbLayer = mpu.get_layerdef_for_proto('DeployData', 'label', None,
**{'ipDims': [1, 16, 2, 1]})
pdef = model.save_caffe_model(ipLayers=[imgLayer, kpLayer, lbLayer], layerOrder=['render1', 'concat1'])
def test_convert():
fName = '/work4/pulkitag-code/code/ief/IEF/models/new_models/models/new-model.mat'
outName = 'try.prototxt'
model = MatConvNetModel(fName)
imgLayer = mpu.get_layerdef_for_proto('DeployData', 'image', None,
**{'ipDims': [1, 3, 224, 224]})
kpLayer = mpu.get_layerdef_for_proto('DeployData', 'kp_pos', None,
**{'ipDims': [1, 17, 2, 1]})
lbLayer = mpu.get_layerdef_for_proto('DeployData', 'label', None,
**{'ipDims': [1, 16, 2, 1]})
pdef = model.save_caffe_model(ipLayers=[imgLayer, kpLayer, lbLayer],
layerOrder=['render1', 'concat1'])
def make_def_proto(nw,
isSiamese=True,
baseFileStr='split_im.prototxt',
getStreamTopNames=False):
'''
If is siamese then wait for the Concat layers - and make all layers until then siamese.
'''
baseFile = os.path.join(baseFileStr)
protoDef = mpu.ProtoDef(baseFile)
#if baseFileStr in ['split_im.prototxt', 'normal.prototxt']:
lastTop = 'data'
siameseFlag = isSiamese
stream1, stream2 = [], []
mainStream = []
nameGen = mpu.LayerNameGenerator()
for l in nw:
lType, lParam = l
lName = nameGen.next_name(lType)
#To account for layers that should not copied while finetuning
# Such layers need to named differently.
if lParam.has_key('nameDiff'):
lName = lName + '-%s' % lParam['nameDiff']
if lType == 'Concat':
siameseFlag = False
if not lParam.has_key('bottom2'):
lParam['bottom2'] = lastTop + '_p'
if siameseFlag:
lDef, lsDef = mpu.get_siamese_layerdef_for_proto(
lType, lName, lastTop, **lParam)
stream1.append(lDef)
stream2.append(lsDef)
else:
lDef = mpu.get_layerdef_for_proto(lType, lName, lastTop, **lParam)
mainStream.append(lDef)
if lParam.has_key('shareBottomWithNext'):
assert lParam['shareBottomWithNext']
pass
else:
lastTop = lName
#Add layers
mainStream = stream1 + stream2 + mainStream
for l in mainStream:
protoDef.add_layer(l['name'][1:-1], l)
if getStreamTopNames:
if isSiamese:
top1Name = stream1[-1]['name'][1:-1]
top2Name = stream2[-1]['name'][1:-1]
else:
top1Name, top2Name = None, None
return protoDef, top1Name, top2Name
else:
return protoDef
def make_def_proto(nw, isSiamese=True,
baseFileStr='split_im.prototxt', getStreamTopNames=False):
'''
If is siamese then wait for the Concat layers - and make all layers until then siamese.
'''
baseFile = os.path.join(baseFileStr)
protoDef = mpu.ProtoDef(baseFile)
#if baseFileStr in ['split_im.prototxt', 'normal.prototxt']:
lastTop = 'data'
siameseFlag = isSiamese
stream1, stream2 = [], []
mainStream = []
nameGen = mpu.LayerNameGenerator()
for l in nw:
lType, lParam = l
lName = nameGen.next_name(lType)
#To account for layers that should not copied while finetuning
# Such layers need to named differently.
if lParam.has_key('nameDiff'):
lName = lName + '-%s' % lParam['nameDiff']
if lType == 'Concat':
siameseFlag = False
if not lParam.has_key('bottom2'):
lParam['bottom2'] = lastTop + '_p'
if siameseFlag:
lDef, lsDef = mpu.get_siamese_layerdef_for_proto(lType, lName, lastTop, **lParam)
stream1.append(lDef)
stream2.append(lsDef)
else:
lDef = mpu.get_layerdef_for_proto(lType, lName, lastTop, **lParam)
mainStream.append(lDef)
if lParam.has_key('shareBottomWithNext'):
assert lParam['shareBottomWithNext']
pass
else:
lastTop = lName
#Add layers
mainStream = stream1 + stream2 + mainStream
for l in mainStream:
protoDef.add_layer(l['name'][1:-1], l)
if getStreamTopNames:
if isSiamese:
top1Name = stream1[-1]['name'][1:-1]
top2Name = stream2[-1]['name'][1:-1]
else:
top1Name, top2Name = None, None
return protoDef, top1Name, top2Name
else:
return protoDef
def setup_experiment_finetune(prms, cPrms, returnTgCPrms=False, srcDefFile=None):
if srcDefFile is None:
#Get the def file.
if cPrms['fine']['extraFc'] and cPrms['fine']['addDrop']:
defFile = os.path.join(baseFilePath,
'kitti_finetune_fc6_drop_extraFc_deploy.prototxt')
else:
if cPrms['concatLayer'] == 'conv4' and cPrms['isMySimple']:
defFile = os.path.join(baseFilePath, 'kitti_finetune_conv4_mysimple_deploy.prototxt')
else:
defFile = os.path.join(baseFilePath,
'kitti_finetune_fc6_deploy.prototxt')
else:
defFile = srcDefFile
#Setup the target experiment.
tgCPrms = get_caffe_prms(isFineTune=True,
convConcat = cPrms['convConcat'],
fine_base_lr=cPrms['fine']['base_lr'],
fineRunNum = cPrms['fine']['runNum'],
sourceModelIter = cPrms['fine']['modelIter'],
lrAbove = cPrms['fine']['lrAbove'],
fineNumData = cPrms['fine']['numData'],
fineMaxLayer = cPrms['fine']['maxLayer'],
fineDataSet = cPrms['fine']['dataset'],
fineMaxIter = cPrms['fine']['max_iter'],
deviceId = cPrms['deviceId'],
addDrop = cPrms['fine']['addDrop'], extraFc = cPrms['fine']['extraFc'],
imgntMean=cPrms['imgntMean'], stepsize=cPrms['stepsize'],
batchSz=cPrms['batchSz'],
concatLayer = cPrms['concatLayer'],
isMySimple = cPrms['isMySimple'],
imSz = cPrms['imSz'],
contextPad = cPrms['contextPad'])
tgPrms = copy.deepcopy(prms)
tgPrms['expName'] = 'fine-FROM-%s' % prms['expName']
tgExp = get_experiment_object(tgPrms, tgCPrms)
tgExp.init_from_external(tgCPrms['solver'], defFile)
print (tgExp.expFile_.netDef_.get_all_layernames())
#pdb.set_trace()
if not tgCPrms['fine']['maxLayer'] is None:
fcLayer = copy.copy(tgExp.expFile_.netDef_.layers_['TRAIN']['class_fc'])
lossLayer = copy.copy(tgExp.expFile_.netDef_.layers_['TRAIN']['loss'])
accLayer = copy.copy(tgExp.expFile_.netDef_.layers_['TRAIN']['accuracy'])
tgExp.del_all_layers_above(tgCPrms['fine']['maxLayer'])
mxLayer = tgCPrms['fine']['maxLayer']
lastTop = tgExp.get_last_top_name()
if tgCPrms['fine']['addDrop']:
dropLayer = mpu.get_layerdef_for_proto('Dropout', 'drop-%s' % lastTop, lastTop,
**{'top': lastTop, 'dropout_ratio': 0.5})
tgExp.add_layer('drop-%s' % lastTop, dropLayer, 'TRAIN')
if tgCPrms['fine']['extraFc']:
eName = 'fc-extra'
ipLayer = mpu.get_layerdef_for_proto('InnerProduct', eName, lastTop,
**{'top': eName, 'num_output': 2048})
reLayer = mpu.get_layerdef_for_proto('ReLU', 'relu-extra', eName, **{'top': eName})
tgExp.add_layer(eName, ipLayer, 'TRAIN')
tgExp.add_layer('relu-extra', reLayer, 'TRAIN')
lastTop = eName
if tgCPrms['fine']['addDrop']:
dropLayer = mpu.get_layerdef_for_proto('Dropout', 'drop-%s' % eName, eName,
**{'top': eName, 'dropout_ratio': 0.5})
tgExp.add_layer('drop-%s' % eName, dropLayer, 'TRAIN')
fcLayer['bottom'] = '"%s"' % lastTop
tgExp.add_layer('class_fc', fcLayer, phase='TRAIN')
tgExp.add_layer('loss', lossLayer, phase='TRAIN')
tgExp.add_layer('accuracy', accLayer, phase='TRAIN')
#Do things as needed.
if not tgCPrms['fine']['lrAbove'] is None:
tgExp.finetune_above(tgCPrms['fine']['lrAbove'])
#Put the right data files.
if tgCPrms['fine']['prms'] is None:
assert (tgCPrms['fine']['numData'] == 1) and (tgCPrms['fine']['dataset']=='sun')
dbPath = '/data0/pulkitag/data_sets/sun/leveldb_store'
dbFile = os.path.join(dbPath, 'sun-leveldb-%s-%d')
trnFile = dbFile % ('train', tgCPrms['fine']['runNum'])
tstFile = dbFile % ('test', tgCPrms['fine']['runNum'])
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LEVELDB', phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LEVELDB', phase='TEST')
else:
trnFile = tgCPrms['fine']['prms']['paths']['lmdb']['train']
tstFile = tgCPrms['fine']['prms']['paths']['lmdb']['test']
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LMDB', phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LMDB', phase='TEST')
#Set the data files
tgExp.set_layer_property('data', ['data_param', 'source'],
'"%s"' % trnFile, phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'source'],
'"%s"' % tstFile, phase='TEST')
#Set the imagenet mean
if cPrms['imgntMean']:
#muFile = '"%s"' % '/data1/pulkitag/caffe_models/ilsvrc2012_mean.binaryproto'
muFile = tgCPrms['fine']['muFile']
print muFile
tgExp.set_layer_property('data', ['transform_param', 'mean_file'], muFile, phase='TRAIN')
tgExp.set_layer_property('data', ['transform_param', 'mean_file'], muFile, phase='TEST')
#Set the batch-size
tgExp.set_layer_property('data', ['data_param', 'batch_size'],
tgCPrms['batchSz'], phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'batch_size'],
tgCPrms['batchSz'], phase='TEST')
if returnTgCPrms:
return tgExp, tgCPrms
else:
return tgExp
def make_caffe_layer(self, lNum):
#Get the name
nameRef = self.dat_['net']['layers']['name'][lNum][0]
name = ou.ints_to_str(self.dat_['#refs#'][nameRef][:])
#Get the inputs
ipRef = self.dat_['net']['layers']['inputs'][lNum][0]
ipNames = self.ref_to_str(ipRef)
#Get the parameter names
pmRef = self.dat_['net']['layers']['params'][lNum][0]
pmNames = self.ref_to_str(pmRef)
#Get the output names
opRef = self.dat_['net']['layers']['outputs'][lNum][0]
opNames = self.ref_to_str(opRef)
#Get the layer type
tpRef = self.dat_['net']['layers']['type'][lNum][0]
lType = ou.ints_to_str(self.dat_['#refs#'][tpRef][:])
#Get the layer params
lpRef = self.dat_['net']['layers']['block'][lNum][0]
lParam = self.dat_['#refs#'][lpRef]
assert (lType[0:5] == 'dagnn')
lType = lType[6:]
if lType == 'Conv':
paramW = {'name': pmNames[0]}
paramB = {'name': pmNames[1]}
pDupKey = mpu.make_key('param', ['param'])
lDef = mpu.get_layerdef_for_proto(
'Convolution', name, ipNames[0], **{
'num_output': int(lParam['size'][3][0]),
'param': paramW,
pDupKey: paramB,
'kernel_size': int(lParam['size'][0][0]),
'stride': int(lParam['stride'][0][0]),
'pad': int(lParam['pad'][0][0])
})
elif lType == 'ReLU':
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'top': opNames[0]})
elif lType == 'Pooling':
poolType = lParam['method'][0]
if poolType == 'max':
poolType = 'MAX'
elif poolType == 'avg':
poolType = 'AVE'
lDef = mpu.get_layerdef_for_proto(
lType, name, ipNames[0], **{
'top': opNames[0],
'kernel_size': int(lParam['poolSize'][0][0]),
'stride': int(lParam['stride'][0][0]),
'pad': int(lParam['pad'][0][0]),
'pool': poolType
})
elif lType == 'LRN':
N, kappa, alpha, beta = lParam['param'][0][0], lParam['param'][1][0],\
lParam['param'][2][0], lParam['param'][3][0]
lDef = mpu.get_layerdef_for_proto(
lType, name, ipNames[0], **{
'top': opNames[0],
'local_size': int(N),
'alpha': N * alpha,
'beta': beta,
'k': kappa
})
elif lType == 'Concat':
lDef = mpu.get_layerdef_for_proto(
lType, name, ipNames[0], **{
'bottom2': ipNames[1:],
'concat_dim': 1,
'top': opNames[0]
})
elif lType == 'Loss':
lossType = ou.ints_to_str(lParam['loss'])
if lossType == 'pdist':
p = lParam['p'][0][0]
if p == 2:
lossName = 'EuclideanLoss'
else:
raise Exception('Loss type %s not recognized' % lossType)
else:
raise Exception('Loss type %s not recognized' % lossType)
lDef = mpu.get_layerdef_for_proto(lossName, name, ipNames[0],
**{'bottom2': ipNames[1]})
elif lType == 'gaussRender':
lDef = mpu.get_layerdef_for_proto(
lType, name, ipNames[0], **{
'top': opNames[0],
'K': lParam['K'][0][0],
'T': lParam['T'][0][0],
'sigma': lParam['sigma'][0][0],
'imgSz': int(lParam['img_size'][0][0])
})
else:
raise Exception('Layer Type %s not recognized, %d' % (lType, lNum))
return lDef
def make_net_proto(prms, cPrms, finePrms=None):
baseFilePath = prms.paths.baseNetsDr
isSiamese = False
if prms.isSiamese:
isSiamese = True
if finePrms is not None and not finePrms.isSiamese:
isSiamese = False
if isSiamese:
netFileStr = '%s_window_siamese_%s.prototxt'
else:
netFileStr = '%s_window_%s.prototxt'
netFile = netFileStr % (cPrms.nwPrms.netName, cPrms.nwPrms.concatLayer)
netFile = osp.join(baseFilePath, netFile)
netDef = mpu.ProtoDef(netFile)
if cPrms.nwPrms.extraFc is not None:
#Changethe name of the existing common_fc to common_fc_prev
netDef.rename_layer('common_fc', 'common_fc_prev')
netDef.set_layer_property('common_fc_prev', 'top',
'"%s"' % 'common_fc_prev')
#Rename the params
netDef.set_layer_property('common_fc_prev', ['param', 'name'],
'"%s"' % 'common_fc_prev_w')
netDef.set_layer_property('common_fc_prev', ['param', 'name'],
'"%s"' % 'common_fc_prev_b',
propNum=[1, 0])
netDef.rename_layer('relu_common', 'relu_common_prev')
netDef.set_layer_property('relu_common_prev', 'top',
'"%s"' % 'common_fc_prev')
netDef.set_layer_property('relu_common_prev', 'bottom',
'"%s"' % 'common_fc_prev')
#Add the new layer
eName = 'common_fc'
lastTop = 'common_fc_prev'
fcLayer = mpu.get_layerdef_for_proto(
'InnerProduct', eName, lastTop, **{
'top': eName,
'num_output': cPrms.nwPrms.extraFc
})
reLayer = mpu.get_layerdef_for_proto('ReLU', 'relu_common', eName,
**{'top': eName})
netDef.add_layer(eName, fcLayer)
netDef.add_layer('relu_common', reLayer)
if cPrms.nwPrms.numFc5 is not None:
netDef.set_layer_property('fc5', ['inner_product_param', 'num_output'],
'%d' % cPrms.nwPrms.numFc5)
if prms.isSiamese:
netDef.set_layer_property('fc5_p',
['inner_product_param', 'num_output'],
'%d' % cPrms.nwPrms.numFc5)
if cPrms.nwPrms.numConv4 is not None:
netDef.set_layer_property('conv4', ['convolution_param', 'num_output'],
'%d' % cPrms.nwPrms.numConv4)
if prms.isSiamese:
netDef.set_layer_property('conv4_p',
['convolution_param', 'num_output'],
'%d' % cPrms.nwPrms.numConv4)
if cPrms.nwPrms.numCommonFc is not None:
netDef.set_layer_property('common_fc',
['inner_product_param', 'num_output'],
'%d' % cPrms.nwPrms.numCommonFc)
if cPrms.nwPrms.concatDrop:
dropLayer = mpu.get_layerdef_for_proto(
'Dropout', 'drop-%s' % 'common_fc', 'common_fc', **{
'top': 'common_fc',
'dropout_ratio': 0.5
})
netDef.add_layer('drop-%s' % 'common_fc', dropLayer, 'TRAIN')
if finePrms is not None:
netDef.rename_layer('common_fc', 'common_fc_fine')
return netDef
def setup_experiment_finetune(prms,
cPrms,
returnTgCPrms=False,
srcDefFile=None):
if srcDefFile is None:
#Get the def file.
if cPrms['fine']['extraFc'] and cPrms['fine']['addDrop']:
defFile = os.path.join(
baseFilePath,
'kitti_finetune_fc6_drop_extraFc_deploy.prototxt')
else:
if cPrms['concatLayer'] == 'conv4' and cPrms['isMySimple']:
defFile = os.path.join(
baseFilePath,
'kitti_finetune_conv4_mysimple_deploy.prototxt')
else:
defFile = os.path.join(baseFilePath,
'kitti_finetune_fc6_deploy.prototxt')
else:
defFile = srcDefFile
#Setup the target experiment.
tgCPrms = get_caffe_prms(isFineTune=True,
convConcat=cPrms['convConcat'],
fine_base_lr=cPrms['fine']['base_lr'],
fineRunNum=cPrms['fine']['runNum'],
sourceModelIter=cPrms['fine']['modelIter'],
lrAbove=cPrms['fine']['lrAbove'],
fineNumData=cPrms['fine']['numData'],
fineMaxLayer=cPrms['fine']['maxLayer'],
fineDataSet=cPrms['fine']['dataset'],
fineMaxIter=cPrms['fine']['max_iter'],
deviceId=cPrms['deviceId'],
addDrop=cPrms['fine']['addDrop'],
extraFc=cPrms['fine']['extraFc'],
imgntMean=cPrms['imgntMean'],
stepsize=cPrms['stepsize'],
batchSz=cPrms['batchSz'],
concatLayer=cPrms['concatLayer'],
isMySimple=cPrms['isMySimple'],
imSz=cPrms['imSz'],
contextPad=cPrms['contextPad'])
tgPrms = copy.deepcopy(prms)
tgPrms['expName'] = 'fine-FROM-%s' % prms['expName']
tgExp = get_experiment_object(tgPrms, tgCPrms)
tgExp.init_from_external(tgCPrms['solver'], defFile)
print(tgExp.expFile_.netDef_.get_all_layernames())
#pdb.set_trace()
if not tgCPrms['fine']['maxLayer'] is None:
fcLayer = copy.copy(
tgExp.expFile_.netDef_.layers_['TRAIN']['class_fc'])
lossLayer = copy.copy(tgExp.expFile_.netDef_.layers_['TRAIN']['loss'])
accLayer = copy.copy(
tgExp.expFile_.netDef_.layers_['TRAIN']['accuracy'])
tgExp.del_all_layers_above(tgCPrms['fine']['maxLayer'])
mxLayer = tgCPrms['fine']['maxLayer']
lastTop = tgExp.get_last_top_name()
if tgCPrms['fine']['addDrop']:
dropLayer = mpu.get_layerdef_for_proto(
'Dropout', 'drop-%s' % lastTop, lastTop, **{
'top': lastTop,
'dropout_ratio': 0.5
})
tgExp.add_layer('drop-%s' % lastTop, dropLayer, 'TRAIN')
if tgCPrms['fine']['extraFc']:
eName = 'fc-extra'
ipLayer = mpu.get_layerdef_for_proto(
'InnerProduct', eName, lastTop, **{
'top': eName,
'num_output': 2048
})
reLayer = mpu.get_layerdef_for_proto('ReLU', 'relu-extra', eName,
**{'top': eName})
tgExp.add_layer(eName, ipLayer, 'TRAIN')
tgExp.add_layer('relu-extra', reLayer, 'TRAIN')
lastTop = eName
if tgCPrms['fine']['addDrop']:
dropLayer = mpu.get_layerdef_for_proto(
'Dropout', 'drop-%s' % eName, eName, **{
'top': eName,
'dropout_ratio': 0.5
})
tgExp.add_layer('drop-%s' % eName, dropLayer, 'TRAIN')
fcLayer['bottom'] = '"%s"' % lastTop
tgExp.add_layer('class_fc', fcLayer, phase='TRAIN')
tgExp.add_layer('loss', lossLayer, phase='TRAIN')
tgExp.add_layer('accuracy', accLayer, phase='TRAIN')
#Do things as needed.
if not tgCPrms['fine']['lrAbove'] is None:
tgExp.finetune_above(tgCPrms['fine']['lrAbove'])
#Put the right data files.
if tgCPrms['fine']['prms'] is None:
assert (tgCPrms['fine']['numData']
== 1) and (tgCPrms['fine']['dataset'] == 'sun')
dbPath = '/data0/pulkitag/data_sets/sun/leveldb_store'
dbFile = os.path.join(dbPath, 'sun-leveldb-%s-%d')
trnFile = dbFile % ('train', tgCPrms['fine']['runNum'])
tstFile = dbFile % ('test', tgCPrms['fine']['runNum'])
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LEVELDB',
phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LEVELDB',
phase='TEST')
else:
trnFile = tgCPrms['fine']['prms']['paths']['lmdb']['train']
tstFile = tgCPrms['fine']['prms']['paths']['lmdb']['test']
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LMDB',
phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'backend'],
'LMDB',
phase='TEST')
#Set the data files
tgExp.set_layer_property('data', ['data_param', 'source'],
'"%s"' % trnFile,
phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'source'],
'"%s"' % tstFile,
phase='TEST')
#Set the imagenet mean
if cPrms['imgntMean']:
#muFile = '"%s"' % '/data1/pulkitag/caffe_models/ilsvrc2012_mean.binaryproto'
muFile = tgCPrms['fine']['muFile']
print muFile
tgExp.set_layer_property('data', ['transform_param', 'mean_file'],
muFile,
phase='TRAIN')
tgExp.set_layer_property('data', ['transform_param', 'mean_file'],
muFile,
phase='TEST')
#Set the batch-size
tgExp.set_layer_property('data', ['data_param', 'batch_size'],
tgCPrms['batchSz'],
phase='TRAIN')
tgExp.set_layer_property('data', ['data_param', 'batch_size'],
tgCPrms['batchSz'],
phase='TEST')
if returnTgCPrms:
return tgExp, tgCPrms
else:
return tgExp
def make_caffe_layer(self, lNum):
#Get the name
nameRef = self.dat_['net']['layers']['name'][lNum][0]
name = ou.ints_to_str(self.dat_['#refs#'][nameRef][:])
#Get the inputs
ipRef = self.dat_['net']['layers']['inputs'][lNum][0]
ipNames = self.ref_to_str(ipRef)
#Get the parameter names
pmRef = self.dat_['net']['layers']['params'][lNum][0]
pmNames = self.ref_to_str(pmRef)
#Get the output names
opRef = self.dat_['net']['layers']['outputs'][lNum][0]
opNames = self.ref_to_str(opRef)
#Get the layer type
tpRef = self.dat_['net']['layers']['type'][lNum][0]
lType = ou.ints_to_str(self.dat_['#refs#'][tpRef][:])
#Get the layer params
lpRef = self.dat_['net']['layers']['block'][lNum][0]
lParam = self.dat_['#refs#'][lpRef]
assert (lType[0:5] == 'dagnn')
lType = lType[6:]
if lType == 'Conv':
paramW = {'name': pmNames[0]}
paramB = {'name': pmNames[1]}
pDupKey = mpu.make_key('param', ['param'])
lDef = mpu.get_layerdef_for_proto('Convolution', name, ipNames[0],
**{'num_output': int(lParam['size'][3][0]),
'param': paramW, pDupKey: paramB,
'kernel_size': int(lParam['size'][0][0]),
'stride': int(lParam['stride'][0][0]),
'pad': int(lParam['pad'][0][0])})
elif lType == 'ReLU':
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'top': opNames[0]})
elif lType == 'Pooling':
poolType = lParam['method'][0]
if poolType == 'max':
poolType = 'MAX'
elif poolType == 'avg':
poolType = 'AVE'
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'top': opNames[0], 'kernel_size': int(lParam['poolSize'][0][0]),
'stride': int(lParam['stride'][0][0]), 'pad': int(lParam['pad'][0][0]),
'pool': poolType})
elif lType == 'LRN':
N, kappa, alpha, beta = lParam['param'][0][0], lParam['param'][1][0],\
lParam['param'][2][0], lParam['param'][3][0]
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'top': opNames[0],
'local_size': int(N),
'alpha': N * alpha,
'beta' : beta,
'k' : kappa})
elif lType == 'Concat':
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'bottom2': ipNames[1:],
'concat_dim': 1,
'top': opNames[0]})
elif lType == 'Loss':
lossType = ou.ints_to_str(lParam['loss'])
if lossType == 'pdist':
p = lParam['p'][0][0]
if p == 2:
lossName = 'EuclideanLoss'
else:
raise Exception('Loss type %s not recognized' % lossType)
else:
raise Exception('Loss type %s not recognized' % lossType)
lDef = mpu.get_layerdef_for_proto(lossName, name, ipNames[0],
**{'bottom2': ipNames[1]})
elif lType == 'gaussRender':
lDef = mpu.get_layerdef_for_proto(lType, name, ipNames[0],
**{'top': opNames[0],
'K': lParam['K'][0][0], 'T': lParam['T'][0][0],
'sigma': lParam['sigma'][0][0], 'imgSz': int(lParam['img_size'][0][0])})
else:
raise Exception('Layer Type %s not recognized, %d' % (lType, lNum))
return lDef
