def main(_): pp.pprint(flags.FLAGS.__flags) cfg_stage_i = config_from_yaml(FLAGS.cfg_stage_I) cfg = config_from_yaml(FLAGS.cfg_stage_II) if not os.path.exists(cfg.CHECKPOINT_DIR): os.makedirs(cfg.CHECKPOINT_DIR) if not os.path.exists(cfg.SAMPLE_DIR): os.makedirs(cfg.SAMPLE_DIR) if not os.path.exists(cfg.LOGS_DIR): os.makedirs(cfg.LOGS_DIR) run_config = tf.ConfigProto() run_config.gpu_options.allow_growth = True datadir = cfg.DATASET_DIR dataset = TextDataset(datadir, 256) filename_test = '%s/test' % datadir dataset.test = dataset.get_data(filename_test) filename_train = '%s/train' % datadir dataset.train = dataset.get_data(filename_train) with tf.Session(config=run_config) as sess: if cfg.EVAL.FLAG: stage_i = ConditionalGanStageI(cfg_stage_i, build_model=False) stage_ii = ConditionalGan(stage_i, cfg, build_model=False) stage_ii_eval = StageIIEval( sess=sess, model=stage_ii, dataset=dataset, cfg=cfg, ) stage_ii_eval.evaluate_inception() elif cfg.TRAIN.FLAG: stage_i = ConditionalGanStageI(cfg_stage_i, build_model=False) stage_ii = ConditionalGan(stage_i, cfg) show_all_variables() stage_ii_trainer = ConditionalGanTrainer( sess=sess, model=stage_ii, dataset=dataset, cfg=cfg, cfg_stage_i=cfg_stage_i, ) stage_ii_trainer.train() else: stage_i = ConditionalGanStageI(cfg_stage_i, build_model=False) stage_ii = ConditionalGan(stage_i, cfg, build_model=False) stage_ii_eval = StageIIVisualizer( sess=sess, model=stage_ii, dataset=dataset, cfg=cfg, ) stage_ii_eval.visualize()
def main(_): pp.pprint(flags.FLAGS.__flags) cfg = config_from_yaml(FLAGS.cfg) if not os.path.exists(cfg.CHECKPOINT_DIR): os.makedirs(cfg.CHECKPOINT_DIR) if not os.path.exists(cfg.LOGS_DIR): os.makedirs(cfg.LOGS_DIR) run_config = tf.ConfigProto() run_config.gpu_options.allow_growth = True datadir = cfg.DATASET_DIR dataset = TextDataset(datadir, 299) # We train inception on the test dataset which contains completely other classes from the train dataset # (used in GAN training). This is needed for a correct evaluation of the Inception/FID score. filename_test = '%s/test' % datadir dataset.test = dataset.get_data(filename_test) with tf.Session(config=run_config) as sess: if cfg.TRAIN.FLAG: stage_i_trainer = InceptionTrainer( sess=sess, dataset=dataset, cfg=cfg, ) stage_i_trainer.train() else: pass
def main(_): pp.pprint(flags.FLAGS.__flags) cfg = config_from_yaml(FLAGS.cfg) if not os.path.exists(cfg.CHECKPOINT_DIR): os.makedirs(cfg.CHECKPOINT_DIR) if not os.path.exists(cfg.SAMPLE_DIR): os.makedirs(cfg.SAMPLE_DIR) if not os.path.exists(cfg.LOGS_DIR): os.makedirs(cfg.LOGS_DIR) run_config = tf.ConfigProto() run_config.gpu_options.allow_growth = True datadir = cfg.DATASET_DIR dataset = TextDataset(datadir, 64) # dataset = TextDataset(datadir, 512) filename_test = '%s/test' % datadir dataset._test = dataset.get_data(filename_test) filename_train = '%s/train' % datadir dataset.train = dataset.get_data(filename_train) with tf.Session(config=run_config) as sess: if cfg.EVAL.FLAG: gancls = GanCls(cfg, build_model=False) eval = GanClsEval( sess=sess, model=gancls, dataset=dataset, cfg=cfg) eval.evaluate_inception() elif cfg.TRAIN.FLAG: gancls = GanCls(cfg) show_all_variables() gancls_trainer = GanClsTrainer( sess=sess, model=gancls, dataset=dataset, cfg=cfg, ) gancls_trainer.train() else: gancls = GanCls(cfg, build_model=False) gancls_visualiser = GanClsVisualizer( sess=sess, model=gancls, dataset=dataset, config=cfg, ) gancls_visualiser.visualize()
def main(_): pp.pprint(flags.FLAGS.__flags) cfg = config_from_yaml(FLAGS.cfg) if not os.path.exists(cfg.CHECKPOINT_DIR): os.makedirs(cfg.CHECKPOINT_DIR) if not os.path.exists(cfg.SAMPLE_DIR): os.makedirs(cfg.SAMPLE_DIR) if not os.path.exists(cfg.LOGS_DIR): os.makedirs(cfg.LOGS_DIR) run_config = tf.ConfigProto() run_config.gpu_options.allow_growth = True datadir = cfg.DATASET_DIR dataset = TextDataset(datadir, 1) filename_test = '%s/test' % datadir dataset._test = dataset.get_data(filename_test) filename_train = '%s/train' % datadir dataset.train = dataset.get_data(filename_train) with tf.Session(config=run_config) as sess: stage_i = ConditionalGan(cfg) show_all_variables() if cfg.TRAIN.FLAG: stage_i_trainer = ConditionalGanTrainer( sess=sess, model=stage_i, dataset=dataset, cfg=cfg, ) stage_i_trainer.train() else: pass
flags = tf.app.flags flags.DEFINE_string( 'cfg', './models/pggan/cfg/flowers.yml', 'Relative path to the config of the model [./models/pggan/cfg/flowers.yml]' ) # flags.DEFINE_string('cfg', './models/pggan/cfg/birds.yml', # 'Relative path to the config of the model [./models/pggan/cfg/birds.yml]') FLAGS = flags.FLAGS if __name__ == "__main__": stage = [1, 2, 3, 4, 5, 6, 7] all_samples = [] cfg = config_from_yaml(FLAGS.cfg) batch_size = 64 z_dim = 128 datadir = cfg.DATASET_DIR dataset = TextDataset(datadir, 64) filename_test = '%s/test' % datadir dataset.test = dataset.get_data(filename_test) filename_train = '%s/train' % datadir dataset.train = dataset.get_data(filename_train) z_sample = np.random.standard_normal((batch_size, z_dim))