def __init__(self, args):
device = args.device
_, _, txt_encoder, _, _ = train_retrieval.load_model(
args.retrieval_model, device)
ckpt_args, _, netG, _, _, _ = train_cookgan.load_model(
args.ckpt_path, device)
netG = netG.eval().to(device)
txt_encoder = txt_encoder.eval().to(device)
imsize = ckpt_args.base_size * (2**(ckpt_args.levels - 1))
train_transform = transforms.Compose([
transforms.Resize(int(imsize * 76 / 64)),
transforms.CenterCrop(imsize)
])
dataset = FoodDataset(recipe_file=ckpt_args.recipe_file,
img_dir=ckpt_args.img_dir,
levels=ckpt_args.levels,
part='val',
food_type=ckpt_args.food_type,
base_size=ckpt_args.base_size,
transform=train_transform)
dataloader = torch.utils.data.DataLoader(dataset,
batch_size=args.batch_size,
num_workers=4)
self.ckpt_args = ckpt_args
self.netG = netG
self.txt_encoder = txt_encoder
self.dataloader = dataloader
self.batch_size = args.batch_size
self.device = device
self.fixed_noise = torch.randn(self.batch_size,
self.ckpt_args.z_dim).to(self.device)
part='train',
food_type=args.food_type,
base_size=args.base_size,
transform=train_transform)
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=args.batch_size,
drop_last=True,
shuffle=True,
num_workers=int(args.workers))
print('train data info:', len(train_set), len(train_loader))
##############################
# model
##############################
ckpt_args, _, txt_encoder, img_encoder, _ = train_retrieval.load_model(
args.retrieval_model, device)
requires_grad(txt_encoder, False)
requires_grad(img_encoder, False)
txt_encoder = txt_encoder.eval()
img_encoder = img_encoder.eval()
if args.ckpt_path:
ckpt_args, batch, netG, optimizerG, netsD, optimizersD = load_model(
args.ckpt_path, device)
wandb_run_id = args.ckpt_path.split('/')[-2]
batch_start = batch + 1
else:
netG, netsD, optimizerG, optimizersD = create_model(args, device)
wandb_run_id = ''
batch_start = 0
# prepare to write
f = open(filename, mode='a')
writer = csv.writer(f, delimiter=',')
ckpt_paths = glob(os.path.join(args.ckpt_dir, '*.ckpt'))
ckpt_paths = sorted(ckpt_paths)
print('records:', ckpt_paths)
print('computed:', computed)
data_loader = None
w2i = None
for ckpt_path in ckpt_paths:
print()
print(f'working on {ckpt_path}')
ckpt_args, _, text_encoder, image_encoder, _ = load_model(
ckpt_path, device)
if not data_loader:
print('loading dataset')
dataset = Dataset(
part='val',
recipe_file=ckpt_args.recipe_file,
img_dir=ckpt_args.img_dir,
word2vec_file=ckpt_args.word2vec_file,
permute_ingrs=ckpt_args.permute_ingrs,
transform=val_transform,
)
w2i = dataset.w2i
dataset = torch.utils.data.Subset(dataset,
indices=np.random.choice(
len(dataset), 5000))