def test_merge_w_channels(self):
X_train, Y_train = get_ds_simple(cnt_samples=10)
X_train = np.expand_dims(X_train, axis=1).astype(np.float32) / 255
Y_train = Y_train[:, np.newaxis]
print(X_train.shape, Y_train.shape)
im = merge_samples(X_train, Y_train)
im.save("/tmp/simple_channel.png")
def test_simple_old(self):
print("generating sample output")
cnt_samples = 10
X_train = np.array([gen_item(i % 2) for i in range(cnt_samples)])
Y_train = np.array([i % 2 for i in range(cnt_samples)], dtype=np.int32)
print(X_train.shape, Y_train.shape)
im = merge_samples(X_train, Y_train)
im.save("/tmp/test_legacy.png")
def main():
net = Net()
model = Classifier(net)
# res = net(X_train[0:1])
# print(res.shape)
# generate examples before training
print("image size : ", X_train[:1].size)
im = merge_samples(X_train[:10], Y_train[:10])
im.save("/tmp/ae_0_original.png")
noisy_X = make_noise(X_train)
im = merge_samples(noisy_X, Y_train[:10])
im.save("/tmp/ae_1_noisy_input.png")
encoded = net.encode(X_train[:1])
print("encoded size", encoded.shape)
generated = net(X_train[:10])
im = merge_samples(generated.data, Y_train)
im.save("/tmp/ae_2_untrained.png")
ds_train = chainer.datasets.tuple_dataset.TupleDataset(noisy_X, X_train)
if len(params["gpus"]) > 0:
chainer.cuda.get_device(0).use()
model.to_gpu()
train(model, ds_train, None, params)
if len(params["gpus"]) > 0:
model.to_cpu()
decoded = net(X_train[:10])
im = merge_samples(decoded.data, Y_train[:10])
im.save("/tmp/ae_3_trained.png")
denoised = net(noisy_X[:10])
print(denoised.data.min(), denoised.data.max(), denoised.data.mean())
im = merge_samples(denoised.data, Y_train[:10])
im.save("/tmp/ae_4_trained_denoised.png")
def test_size(self):
X_train, Y_train = get_ds_simple(dim_image=128, cnt_samples=10)
print(X_train.shape, Y_train.shape)
im = merge_samples(X_train, Y_train)
im.save("/tmp/size128.png")
def test_naive(self):
X_train, Y_train = get_ds_naive(cnt_samples=10)
print(X_train.shape, Y_train.shape)
im = merge_samples(X_train, Y_train)
im.save("/tmp/naive.png")
def test_counting(self):
X_train, Y_train = get_ds_counting()
print(X_train.shape, Y_train.shape)
im = merge_samples(X_train, Y_train)
im.save("/tmp/counting.png")