import torch.nn as nn import numpy as np from library import inputs from Utils.checkpoints import save_context, Logger from Utils import flags from Utils import config import Torture from library import loss_triplegan, evaluation import library.loss_cla as loss_classifier from library.mean_teacher import optim_weight_swa FLAGS = flags.FLAGS KEY_ARGUMENTS = config.load_config(FLAGS.config_file) text_logger, MODELS_FOLDER, SUMMARIES_FOLDER = save_context( __file__, KEY_ARGUMENTS) # FLAGS.g_model_name = FLAGS.model_name # FLAGS.d_model_name = FLAGS.model_name torch.manual_seed(1234) torch.cuda.manual_seed(1235) np.random.seed(1236) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = True FLAGS.device = device = torch.device( "cuda" if torch.cuda.is_available() else "cpu") n_iter_d = 5 if "sngan" in FLAGS.g_model_name else 1 def sigmoid_rampup(global_step, start_iter, end_iter):
flags.DEFINE_boolean("-combine_with_JPEG", default=False) flags.DEFINE_boolean("-combine_with_Grayscale", default=False) flags.DEFINE_argument("-Raffprobability", type=float, default=0.2) flags.DEFINE_argument("-JPEGquality", type=int, default=75) flags.DEFINE_argument("-eps", type=float, default=8.) flags.DEFINE_argument("-numtest", type=int, default=50000) FLAGS = flags.FLAGS logger, MODELS_FOLDER, SUMMARIES_FOLDER = save_context(__file__, FLAGS, config) logger.info("build dataloader") num_test = FLAGS.numtest num_pool = 10000 # num of times to sample for MI num_sample = FLAGS.num_sample def onehot(ind, num_cla): vector = np.zeros([num_cla]) vector[ind] = 1 return vector.astype(np.float32) eps_ = 2. clip_min, clip_max = -1., 1.
from Torture.utils import distributions from Torture import shortcuts from ESM.models import Res18_Quadratic from ESM.data import inf_train_gen_cifar FILES_TO_BE_SAVED = ["./", './configs', './ESM'] KEY_ARGUMENTS = ['data', 'esm_eps', 'esm_type'] CONFIG = { "FILES_TO_BE_SAVED": FILES_TO_BE_SAVED, "KEY_ARGUMENTS": KEY_ARGUMENTS } FLAGS = flags.FLAGS config.load_config(FLAGS.config_file) text_logger, MODELS_FOLDER, SUMMARIES_FOLDER = save_context( __file__, FLAGS, CONFIG) shutil.copy(FLAGS.config_file, os.path.join(SUMMARIES_FOLDER, "config.yaml")) torch.manual_seed(1234) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False np.random.seed(12345) device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') model = Res18_Quadratic(3, FLAGS.n_chan, 32).to(device) optimizer = torch.optim.Adam(model.parameters(), 5e-5) checkpoint_io = Torture.utils.checkpoint.CheckpointIO( checkpoint_dir=MODELS_FOLDER) checkpoint_io.register_modules(model=model) logger = Logger(log_dir=SUMMARIES_FOLDER)