def main():
if len(sys.argv) < 7:
print(
'Usage: python3 sample.py [MODEL_PATH] [DATA_PATH] [SPLIT_INDEX] [OUT_PATH] [RECIPE_ID] [NUM_SAMPLES]'
)
exit()
model_path = os.path.abspath(sys.argv[1])
data_path = os.path.abspath(sys.argv[2])
split_index = int(sys.argv[3])
out_path = os.path.abspath(sys.argv[4])
recipe_id = sys.argv[5]
num_samples = int(sys.argv[6])
util.create_dir(out_path)
saved_model = torch.load(model_path)
data = GANstronomyDataset(data_path, split=opts.TVT_SPLIT)
data.set_split_index(split_index)
data_loader = DataLoader(data,
batch_size=1,
shuffle=False,
sampler=SequentialSampler(data))
num_classes = data.num_classes()
G = Generator(opts.LATENT_SIZE, opts.EMBED_SIZE).to(opts.DEVICE)
G.load_state_dict(saved_model['G_state_dict'])
G.eval()
embs = None
for ibatch, data_batch in enumerate(data_loader):
with torch.no_grad():
recipe_ids, recipe_embs, img_ids, imgs, classes, _, _ = data_batch
batch_size, recipe_embs, imgs = util.get_variables3(
recipe_ids, recipe_embs, img_ids, imgs)
if recipe_ids[0] == recipe_id:
embs = recipe_embs
break
assert embs is not None
z = torch.randn(num_samples, opts.LATENT_SIZE).to(opts.DEVICE)
imgs_gen = G(z, embs.expand(num_samples, opts.EMBED_SIZE)).detach()
for i in range(num_samples):
img_gen = imgs_gen[i]
save_results(out_path, recipe_id, img_gen, i)
def main():
if len(sys.argv) < 7:
print('Usage: python3 interp.py [MODEL_PATH] [DATA_PATH] [SPLIT_INDEX] [OUT_PATH] [RECIPE_ID] [NUM_DIV]')
exit()
model_path = os.path.abspath(sys.argv[1])
data_path = os.path.abspath(sys.argv[2])
split_index = int(sys.argv[3])
out_path = os.path.abspath(sys.argv[4])
recipe_id = sys.argv[5]
num_div = int(sys.argv[6])
util.create_dir(out_path)
saved_model = torch.load(model_path)
data = GANstronomyDataset(data_path, split=opts.TVT_SPLIT)
data.set_split_index(split_index)
data_loader = DataLoader(data, batch_size=1, shuffle=False, sampler=SequentialSampler(data))
num_classes = data.num_classes()
G = Generator(opts.LATENT_SIZE, opts.EMBED_SIZE).to(opts.DEVICE)
G.load_state_dict(saved_model['G_state_dict'])
G.eval()
embs = None
for ibatch, data_batch in enumerate(data_loader):
with torch.no_grad():
recipe_ids, recipe_embs, img_ids, imgs, classes, _, _ = data_batch
batch_size, recipe_embs, imgs = util.get_variables3(recipe_ids, recipe_embs, img_ids, imgs)
if recipe_ids[0] == recipe_id:
embs = recipe_embs
break
assert embs is not None
z1 = torch.randn(1, opts.LATENT_SIZE).to(opts.DEVICE)
z2 = torch.randn(1, opts.LATENT_SIZE).to(opts.DEVICE)
for a in np.linspace(0.0, 1.0, num_div + 1):
a = torch.tensor(a, dtype=torch.float)
a = Variable(a.type(FloatTensor)).to(opts.DEVICE)
z = (1.0 - a) * z1 + a * z2
img_gen = G(z, embs).detach()[0]
save_results(out_path, recipe_id, img_gen, a)
def main():
if len(sys.argv) < 5:
print(
'Usage: python3 test.py [MODEL_PATH] [DATA_PATH] [SPLIT_INDEX] [OUT_PATH]'
)
exit()
model_path = os.path.abspath(sys.argv[1])
data_path = os.path.abspath(sys.argv[2])
split_index = int(sys.argv[3])
out_path = os.path.abspath(sys.argv[4])
util.create_dir(out_path)
saved_model = torch.load(model_path)
data = GANstronomyDataset(data_path, split=opts.TVT_SPLIT)
data.set_split_index(split_index)
data_loader = DataLoader(data,
batch_size=opts.BATCH_SIZE,
shuffle=False,
sampler=SequentialSampler(data))
G = Generator(opts.LATENT_SIZE, opts.EMBED_SIZE).to(opts.DEVICE)
G.load_state_dict(saved_model['G_state_dict'])
G.eval()
all_ingrs = util.load_ingredients()
for ibatch, data_batch in enumerate(data_loader):
with torch.no_grad():
recipe_ids, recipe_embs, img_ids, imgs, classes, _, _ = data_batch
batch_size, recipe_embs, imgs, = util.get_variables3(
recipe_ids, recipe_embs, img_ids, imgs)
z = torch.randn(batch_size, opts.LATENT_SIZE).to(opts.DEVICE)
imgs_gen = G(z, recipe_embs)
imgs, imgs_gen = imgs.detach(), imgs_gen.detach()
for iexample in range(batch_size):
save_results(all_ingrs, imgs[iexample], imgs_gen[iexample],
img_ids[iexample], recipe_ids[iexample], out_path)