def main(args):
dataset = VidDataSet(K=args.K,
path_to_mp4=args.data_dir,
device=torch.device("cuda"),
join_by_video=True)
print("Dataset size", len(dataset))
for idx in tqdm.tqdm(range(args.start_idx, args.end_idx + 1)):
filename = create_filename(*dataset.get_video_info(idx))
output_path = os.path.join(args.output_dir, filename)
frame_mark = dataset.get_frame_mark_numpy_array(idx)
np.savez_compressed(output_path, frame_mark=frame_mark)
from network.blocks import *
from network.model import *
from tensorboard_logger import configure, log_value
import pdb
"""Create dataset and net"""
device = torch.device("cuda:0")
cpu = torch.device("cpu")
tensorboard_path = './experiment/tensorboard'
path_to_chkpt = './experiment/model_weights_self_train.tar'
path_to_backup = './experiment/backup_model_weights.tar'
path_to_mp4 = "/home/cxu-serve/p1/common/voxceleb/test/video/sample"
VGGFace_body_path = '/home/cxu-serve/p1/common/vggface/new/Pytorch_VGGFACE_IR.py'
VGGFace_weight_path = '/home/cxu-serve/p1/common/vggface/new/Pytorch_VGGFACE.pth'
dataset = VidDataSet(K=8, path_to_mp4=path_to_mp4, device=device)
dataLoader = DataLoader(dataset, batch_size=2, shuffle=True)
G = Generator(224).to(device)
E = Embedder(224).to(device)
D = Discriminator(dataset.__len__()).to(device)
G.train()
E.train()
D.train()
optimizerG = optim.Adam(params=list(E.parameters()) + list(G.parameters()),
lr=5e-5)
optimizerD = optim.Adam(params=D.parameters(), lr=2e-4)
"""Criterion"""
from network.blocks import *
from network.model import *
from tensorboard_logger import configure, log_value
import pdb
"""Create dataset and net"""
device = torch.device("cuda:0")
cpu = torch.device("cpu")
tensorboard_path = './experiment/tensorboard'
path_to_chkpt = './experiment/model_weights_self_train.tar'
path_to_backup = './experiment/backup_model_weights.tar'
path_to_mp4 = "/home/cxu-serve/p1/common/voxceleb/test/video/sample_one"
VGGFace_body_path = '/home/cxu-serve/p1/common/vggface/new/Pytorch_VGGFACE_IR.py'
VGGFace_weight_path = '/home/cxu-serve/p1/common/vggface/new/Pytorch_VGGFACE.pth'
dataset = VidDataSet(K=8, path_to_mp4=path_to_mp4, device=device)
dataLoader = DataLoader(dataset, batch_size=1, shuffle=True)
G = Generator(224).to(device)
E = Embedder(224).to(device)
# D = Discriminator(dataset.__len__())
# D = Discriminator(dataset.__len__()).to(device)
G.train()
E.train()
# D.train()
optimizerG = optim.Adam(params=list(E.parameters()) + list(G.parameters()),
lr=5e-5)
# optimizerD = optim.Adam(params = D.parameters(), lr=2e-4)
import torchvision.utils as vutils
from dataset.dataset_class import VidDataSet
from dataset.video_extraction_conversion import *
from loss.loss_discriminator import *
from loss.loss_generator import *
from network.blocks import *
from network.model import *
"""Create dataset and net"""
os.environ['CUDA_VISIBLE_DEVICES'] = '1,0'
device = torch.device("cuda")
cpu = torch.device("cpu")
path_to_chkpt = 'model_weights.tar'
path_to_backup = 'backup_model_weights.tar'
dataset = VidDataSet(
K=8,
path_to_mp4='/data2/hao66/dataset/voxceleb1/unzippedFaces',
device=device)
print('# of videos: ', len(dataset))
dataLoader = DataLoader(dataset, batch_size=2, shuffle=True)
G = torch.nn.DataParallel(Generator(224)).to(device)
E = torch.nn.DataParallel(Embedder(224)).to(device)
D = torch.nn.DataParallel(Discriminator(dataset.__len__())).to(device)
G.train()
E.train()
D.train()
optimizerG = optim.Adam(params=list(E.parameters()) + list(G.parameters()),
lr=5e-5)