my_bs =16 kwargs = {'model_name': 'PGAN', 'no_vis': False, 'np_vis': True, 'restart': False, 'name': 'test4', 'dir': '../output_networks', 'configPath': 'config_test.json', 'saveIter': 16000, 'evalIter': 100, 'Scale_iter': None, 'partition_value': None, 'maxIterAtScale': my_maxIterAtScale, 'alphaJumpMode': 'linear', 'iterAlphaJump': None, 'alphaJumpVals': None, 'alphaNJumps': my_alphaNJumps, 'alphaSizeJumps': my_alphaSizeJumps, 'depthScales': None, 'miniBatchSize': my_bs, 'dimLatentVector': None, 'initBiasToZero': None, 'perChannelNormalization': None, 'lossMode': None, 'lambdaGP': None, 'leakyness': None, 'epsilonD': None, 'miniBatchStdDev': None, 'baseLearningRate': None, 'dimOutput': None, 'weightConditionG': None, 'weightConditionD': None, 'GDPP': None, 'overrides': False} # the nEpochs isn't clear what it is # see ./models/trainer/standard_configurations/dcgan_config.py:50:_C.nEpoch = 10 trainingConfig = {'config': {}} #arguments for GANTrainer = trainerModule 'visualisation':need to import vis_module = importlib.import_module("visualization.np_visualizer"),'lossIterEvaluation':100,'checkPointDir':../output_networks/test1,'saveIter':16000,'modelLabel':test1,'partitionValue':None # Build the output durectory if necessary if not kwargs.get('dir','.'): os.mkdir(kwargs.get('dir','.')) # Checkpoint data modelLabel = kwargs["name"] restart = kwargs["restart"] checkPointDir = os.path.join(kwargs["dir"], modelLabel) checkPointData = getLastCheckPoint(checkPointDir, modelLabel) if not os.path.isdir(checkPointDir): os.mkdir(checkPointDir) with open(kwargs["configPath"], 'rb') as file: trainingConfig = json.load(file) trainingConfig['pathDB'] = '../../../data/' trainingConfig['imagefolderDataset'] = True # Model configuration configOverride = getConfigOverrideFromParser(kwargs, trainerModule._defaultConfig) modelConfig = trainingConfig.get("config", {}) for item, val in configOverride.items(): modelConfig[item] = val
# Add overrides to the parser: changes to the model configuration can be # done via the command line parser = updateParserWithConfig(parser, trainerModule._defaultConfig) kwargs = vars(parser.parse_args()) configOverride = getConfigOverrideFromParser( kwargs, trainerModule._defaultConfig) if kwargs['overrides']: parser.print_help() sys.exit() # Checkpoint data modelLabel = kwargs["name"] restart = kwargs["restart"] checkPointDir = os.path.join(kwargs["dir"], modelLabel) checkPointData = getLastCheckPoint(checkPointDir, modelLabel) # models/utils/utils.py: return # checkPointData = trainConfig, pathModel, if not os.path.isdir(checkPointDir): os.mkdir(checkPointDir) # Training configuration configPath = kwargs.get("configPath", None) if configPath is None: raise ValueError("You need to input a configuratrion file") with open(kwargs["configPath"], 'rb') as file: trainingConfig = json.load(file) # Model configuration modelConfig = trainingConfig.get("config", {})