def renderAndSaveModelImageFromJson(paramFlowJson, fout, jobDir=None):
tmpParser = DLSDesignerFlowsParser(paramFlowJson)
if jobDir is not None:
if not os.path.isdir(jobDir):
jobDir = None
tmpModel, _ = tmpParser.buildKerasTrainer(pathJobDir=jobDir)
kplot(tmpModel.model, to_file=fout, show_shapes=True)
return fout
# In[ ]: #genTrainer = GeneratorTrainer(tracker, generator, processor, validator) #Avoid maximum recursion limit exception when pickling by increasing limit from ~1000 by default sys.setrecursionlimit(10000) #genTrainer.train(7, 5, batchSize, lnr, lnrdy) tracker.train(trainGenerator(), offEpochs, offBatches, batchSize, validator) tracker.save(trackerModelPath) # In[13]: #tracker.load(trackerModelPath) from keras.utils.visualize_util import plot as kplot kplot(regressor.model.layer, to_file='/home/fhdiaze/model.png') # # TRACKER TESTING # ## VARIABLES # In[14]: fps = 15 outDir = "/home/fhdiaze/Data/Videos/" # ## VOT TESTING # In[15]: tracker.setStateful(True, 1)
print(t1)
print(t2)
print(t3)
print(t4)
Test.printTest()
####################################
if __name__ == '__main__':
foutJson = 'keras-model-generated.json'
fnFlowJson = '../../../data-test/test-models-json/testnet_multi_input_multi_output_v1.json'
flowParser = DLSDesignerFlowsParserV2(fnFlowJson)
flowParser.cleanAndValidate()
# (1) Build connected and validated Model Node-flow (DLS-model-representation)
flowParser.buildConnectedFlow()
# (2) Generate dict-based Json Kearas model (from DLS model representation)
modelJson = flowParser.generateModelKerasConfigJson()
# (3) Export generated json model to file
with open(foutJson, 'w') as f:
f.write(json.dumps(modelJson, indent=4))
# (4) Try to load generated Keras model from json-file
with open(foutJson, 'r') as f:
model = keras.models.model_from_json(f.read())
# (5) Visualize & summary of the model: check connections!
fimgModel = '%s-figure.jpg' % foutJson
kplot(model, fimgModel, show_shapes=True)
plt.imshow(skio.imread(fimgModel))
plt.grid(True)
plt.show()
model.summary()
def plot(self):
filename = join(self.igor.model_location, self.igor.saving_prefix,
'model_visualization.png')
kplot(self.model, to_file=filename)
self.logger.debug("+ Model visualized at {}".format(filename))
# return os.path.basename(fout)
@staticmethod
def renderModelImageToTmpFile(paramFlowJson,
odir=None,
pref='netgraph_',
jobDir=None):
if odir is not None:
if not os.path.isdir(odir):
os.mkdir(odir)
fout = tempfile.mktemp(prefix=pref, suffix='.png', dir=odir)
tret = DLSDesignerFlowsParser.renderAndSaveModelImageFromJson(
paramFlowJson, fout, jobDir)
return tret
####################################
if __name__ == '__main__':
fnFlowJson = '../../../data-test/test-models-json/test_cnn1.json'
fnFlowJsonOut = '../../../data-test/test-models-json/test_cnn1_out.json'
dirJobs = '../../../data-test/test-models-json/'
flowParser = DLSDesignerFlowsParser(fnFlowJson)
flowParser.cleanAndValidate()
# flowParser.exportConfigFlow(fnFlowJsonOut)
modelTrainer, modelConfig = flowParser.buildKerasTrainer(
pathJobDir=dirJobs)
modelTrainer.saveModelState('/tmp/', isSaveWeights=False)
modelJson = flowParser.buildKerasModelInJson()
kplot(modelTrainer.model, to_file='/tmp/keras_draw.png', show_shapes=True)
print('---------------')
print(json.dumps(modelJson, indent=4))
# pathTestModel='../../../data-test/test-models-json/test_basic_cnn_network_v1.json'
pathTestModel='../../../data-test/test-models-json/test_basic_cnn_network_v1_with_train_params_v1.json'
if __name__ == '__main__':
dirData = dlsutils.getPathForDatasetDir()
dirModels = dlsutils.getPathForModelsDir()
dbWatcher = DatasetsWatcher(dirData)
dbWatcher.refreshDatasetsInfo()
assert ( len(dbWatcher.dictDbInfo.keys())>0 )
dbInfoTest = dbWatcher.dictDbInfo[dbWatcher.dictDbInfo.keys()[0]]
print ('Dataset for tests : [ %s ]' % dbInfoTest.__str__())
#
with open(pathTestModel, 'r') as f:
jsonModelData = json.load(f)
modelParser = DLSDesignerFlowsParser(jsonModelData)
modelTrainer, modelConfig = modelParser.buildKerasTrainer()
batcherDB = BatcherImage2DLMDB(dbInfoTest.pathDB)
#
modelTrainerAdjusted = modelTrainer.adjustModelInputOutput2DBData(modelTrainer.model, batcherDB)
for ii,ll in enumerate(modelTrainerAdjusted.layers):
print ('[%d/%d] : %s, shape: inp=%s, out=%s' % (ii,len(modelTrainerAdjusted.layers), ll, ll.input_shape, ll.output_shape))
print ('*** Total Model params: %d' % modelTrainerAdjusted.count_params())
#
fimg = '/tmp/keras_draw.png'
kplot(modelTrainerAdjusted, to_file=fimg, show_shapes=True)
img = io.imread(fimg)
plt.imshow(img)
plt.show()
#!/usr/bin/python
# -*- coding: utf-8 -*-
__author__ = 'ar'
import skimage.io as io
import matplotlib.pyplot as plt
from keras.utils.visualize_util import plot as kplot
from app.backend.core.models.flow_parser import DLSDesignerFlowsParser
if __name__ == '__main__':
pathModelJson = '../../../data-test/test-models-json/test_cnn1.json'
flowParser = DLSDesignerFlowsParser(pathModelJson)
flowParser.cleanAndValidate()
print('Model-flow isOk: [%s]' % flowParser.isOk())
kerasTrainer, cfgSolver, layersDict = flowParser.buildKerasTrainer(
isPrecalculateLayersDict=True)
print(kerasTrainer.model)
#
fimg = '/tmp/keras_draw.png'
kplot(kerasTrainer.model, to_file=fimg, show_shapes=True)
img = io.imread(fimg)
plt.imshow(img)
plt.show()
pathProto = pp['proto']
pathWeights = pp['weights']
# (1) convert Caffe->Keras
pathKerasModelOutput = '%s-kerasmodel.json' % os.path.splitext(
pathProto)[0]
print('[%d/%d] %s --> %s' %
(ii, len(lstModelsPaths), pathProto, pathKerasModelOutput))
try:
model = caffeConvert.caffe_to_keras(pathProto,
caffemodelPath=None,
debug=False)
with open(pathKerasModelOutput, 'w') as f:
f.write(model.to_json(indent=4))
# (2) try to load and plot Keras model
with open(pathKerasModelOutput, 'r') as f:
jsonData = f.read()
kerasModel = keras.models.model_from_json(
jsonData, custom_objects=dictExtraLayers)
pathKerasModelImage = '%s-kerasmodel.jpg' % os.path.splitext(
pathProto)[0]
kplot(kerasModel, to_file=pathKerasModelImage, show_shapes=True)
img = io.imread(pathKerasModelImage)
plt.subplot(1, numProto, ii + 1)
plt.imshow(img)
plt.title(os.path.splitext(os.path.basename(pathProto))[0])
except Exception as err:
print(
'\t**ERROR** Cant convert and visualize model [%s] : %s, skip...'
% (pathProto, err))
plt.show()
