def evaluation(trainer):
model = load_inception_model()
ims = []
xp = gen.xp
n_ims = 50000
for i in range(0, n_ims, batchsize):
z = Variable(xp.asarray(gen.make_hidden(batchsize)))
with chainer.using_config('train', False), chainer.using_config('enable_backprop', False):
x = gen(z)
x = chainer.cuda.to_cpu(x.data)
x = np.asarray(np.clip(x * 127.5 + 127.5, 0.0, 255.0), dtype=np.uint8)
ims.append(x)
ims = np.asarray(ims)
_, _, _, h, w = ims.shape
ims = ims.reshape((n_ims, 3, h, w)).astype("f")
mean, std = inception_score(model, ims)
chainer.reporter.report({
'inception_mean': mean,
'inception_std': std
})
isarray.append([mean,std])
_IS_array = np.array(isarray)
_IS_array = _IS_array[:,0]
plt.plot(range(len(_IS_array)), _IS_array, 'r-', label = 'Inception Score')
plt.legend()
plt.savefig(dir+'/inception_score.pdf', bbox_inches='tight',format="pdf", dpi = 300)
plt.close()
np.save(dir+"/inception_score.npy",{"IS_array":isarray})
def sample_image(model, encoder, output_image_dir, n_row, batches_done,
dataloader, device):
"""Saves a grid of generated imagenet pictures with captions"""
target_dir = os.path.join(output_image_dir, "samples/")
if not os.path.isdir(target_dir):
os.makedirs(target_dir)
captions = []
gen_imgs = []
# get sample captions
done = False
while not done:
for (_, labels_batch, captions_batch) in dataloader:
captions += captions_batch
conditional_embeddings = encoder(labels_batch.to(device), captions)
imgs = model.sample(conditional_embeddings).cpu()
gen_imgs.append(imgs)
if len(captions) > n_row**2:
done = True
break
break
gen_imgs = torch.cat(gen_imgs).numpy()
gen_imgs = np.clip(gen_imgs + 1e-5, 0, 1)
score = inception_score(gen_imgs, splits=1)
print("Inception score:", score)
for (imgs, labels, captions) in dataloader:
fid = calculate_fid(imgs,
gen_imgs,
batch_size=len(gen_imgs),
use_multiprocessing=False)
break
print("FID score:", fid)
fig = plt.figure(figsize=((8, 8)))
grid = ImageGrid(fig, 111, nrows_ncols=(n_row, n_row), axes_pad=0.2)
for i in range(n_row**2):
grid[i].imshow(gen_imgs[i].transpose([1, 2, 0]))
grid[i].set_title(captions[i])
grid[i].tick_params(axis='both',
which='both',
bottom=False,
top=False,
labelbottom=True)
save_file = os.path.join(target_dir, "{:013d}.png".format(batches_done))
plt.savefig(save_file)
print("saved {}".format(save_file))
plt.close()
def compute_is():
fake_imgs_path = os.path.join(output_dir, name, 'images',
'epoch_{}'.format(epoch))
mean, std = inception_score(fake_imgs_path)
results[name][epoch]['IS'] = {
'mean': '{:.5f}'.format(mean),
'std': '{:.5f}'.format(std)
}
print('IS: ', mean, std)
def my_inception(gen, batchsize=100):
model = load_inception_model()
ims = []
n_ims = 100
for i in range(0, n_ims, batchsize):
z = Variable(np.asarray(gen.make_hidden(batchsize)))
# with chainer.using_config('train', False), chainer.using_config('enable_backprop', False):
# x = gen(z)
x = gen(z).data
# x = chainer.cuda.to_cpu(x.data)
x = np.asarray(np.clip(x * 127.5 + 127.5, 0.0, 255.0), dtype=np.uint8)
ims.append(x)
ims = np.asarray(ims)
_, _, _, h, w = ims.shape
ims = ims.reshape((n_ims, 3, h, w)).astype("f")
mean, std = inception_score(model, ims) #FIXME ここで落ちる
return mean, std
def evaluation(trainer):
model = load_inception_model()
ims = []
xp = gen.xp
n_ims = 50000
for i in range(0, n_ims, batchsize):
z = Variable(xp.asarray(gen.make_hidden(batchsize)))
with chainer.using_config('train', False), chainer.using_config(
'enable_backprop', False):
x = gen(z)
x = chainer.cuda.to_cpu(x.data)
x = np.asarray(np.clip(x * 127.5 + 127.5, 0.0, 255.0),
dtype=np.uint8)
ims.append(x)
ims = np.asarray(ims)
_, _, _, h, w = ims.shape
ims = ims.reshape((n_ims, 3, h, w)).astype("f")
mean, std = inception_score(model, ims) #FIXME ここで落ちる
chainer.reporter.report({'inception_mean': mean, 'inception_std': std})