import os, sys
import tensorflow as tf
from AnetLib.options.train_options import Options
from AnetLib.models.models import create_model
from smlm_datasets import create_data_sources
default_workdir = './output/' + os.path.basename(sys.argv[0])
opt = Options().parse()
opt.fineSize = 512
opt.batchSize = 1 # batchSize = 1
opt.model = 'a_net_tensorflow'
opt.dim_ordering = 'channels_last'
opt.display_freq = 500
opt.save_latest_freq = 1000
opt.use_resize_conv = True
opt.norm_A = 'mean_std'
opt.norm_B = 'min_max[0,1]'
opt.lambda_A = 50
opt.input_nc = 2
opt.lr_nc = 1
opt.lr_scale = 1.0/4.0
opt.lambda_LR = 25
opt.control_nc = 1
opt.add_data_type_control = True
opt.add_lr_channel = False
opt.use_random_channel_mask = True
opt.lr_loss_mode = 'lr_predict'
if opt.phase == 'train':
sources = create_data_sources('TransformedCSVImages', opt)
d = sources['train']
def test_training():
opt = Options().parse(['--workdir=./__test_tmp__/'])
opt.model = 'a_net_tensorflow'
opt.fineSize = 256
opt.batchSize = 1
opt.dim_ordering = 'channels_last'
opt.display_freq = 500
opt.use_resize_conv = True
opt.norm_A = 'mean_std'
opt.norm_B = 'min_max[0,1]'
opt.lambda_A = 50
opt.input_nc = 2
opt.lr_nc = 1
opt.lr_scale = 1.0 / 4.0
opt.lambda_LR = 0
opt.control_nc = 1
opt.add_data_type_control = True
opt.add_lr_channel = 'pseudo'
# reduce the anet size
opt.ngf = 1
opt.ndf = 1
# opt.continue_train = True
# start training
sources = create_data_sources(['TransformedTubulin001NB'], opt)
d = sources['train']
model = create_model(opt)
model.train(d, verbose=1, max_steps=1)
# training done
opt.phase = 'test'
model = create_model(opt)
sources = create_data_sources(['TransformedTubulin001NB'], opt)
d = sources['test']
model.predict(d, verbose=1, max_steps=1)
#!/usr/bin/python # -*- coding: utf-8 -*- ''' Freeze A-Net models python3 freeze.py --workdir=./results/frozen_model_1 --load_dir=./results/simulated_model ''' import os import sys import tensorflow as tf from AnetLib.options.train_options import Options from AnetLib.models.models import create_model from smlm_datasets import create_data_sources from AnetLib.util.freeze_graph import freeze_latest_checkpoint default_workdir = './workdir' opt = Options().parse() opt.model = 'anet_tensorflow' opt.fineSize = 512 opt.batchSize = 1 opt.dim_ordering = 'channels_last' opt.display_freq = 500 opt.use_resize_conv = True opt.norm_A = 'mean_std' opt.norm_B = 'min_max[0,1]' opt.lambda_A = 50 opt.input_nc = 2 opt.lr_nc = 1 opt.lr_scale = 1.0/4.0 opt.lambda_LR = 0 opt.control_nc = 1 opt.add_data_type_control = True