def test2(self):
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, batch_size=32)
n_iters = 0
# dataset_iter = iter(loader)
for batch in iter(loader):
if batch == StopIteration:
break
batch_X, batch_y = batch
logging.debug("len batch {}".format(len(batch_y)))
logging.debug("batch shape {}".format(batch_X.shape))
n_iters += 1
logging.debug("n_iters actual {} | {}".format(n_iters, len(loader)))
def test1_1(self):
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, 32, test=True)
testing_samples, labels, testing_ori, labels_ori = \
loader.X, loader.y, loader.X_ori, loader.y_ori
logging.debug("t_pics {} t_labels {}".format(type(testing_samples),
type(labels)))
logging.debug("n_pics {} n_labels {}".format(len(testing_samples), 0))
n_samples = len(testing_samples)
sampled_ind = random.randint(0, n_samples)
logging.debug("showing picture {}".format(sampled_ind))
# ds.plot_sample(training_ori[sampled_ind], labels_ori[sampled_ind])
ds.plot_testing_sample(testing_samples[sampled_ind])
def test5(self):
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, batch_size=32)
i: int = 0
for i in tqdm.tqdm(range(10000000),
desc="{}train_loop iter {}{}".format(
Fore.RED, i, Style.RESET_ALL)):
for batch in tqdm.tqdm(iter(loader),
desc="{}epoch {}{}".format(
Fore.BLUE, i, Style.RESET_ALL)):
if batch == StopIteration:
# logging.info("{}done with epoch {}{}".format(
# Back.RED, i, Style.RESET_ALL))
logging.info(green("dont with epoch {}".format(cyan(i))))
break
time.sleep(0.02)
def test1(self):
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, 32)
training_samples, labels, training_ori, labels_ori = loader.load()
logging.debug("t_pics {} t_labels {}".format(type(training_samples),
type(labels)))
logging.debug("n_pics {} n_labels {}".format(len(training_samples),
len(labels)))
n_samples = len(labels)
sampled_ind = random.randint(0, n_samples)
logging.debug("showing picture {}".format(sampled_ind))
# ds.plot_sample(training_ori[sampled_ind], labels_ori[sampled_ind])
ds.plot_sample_denormed(training_samples[sampled_ind],
labels[sampled_ind])
def test4(self):
logging.debug("testing network 1 for " "regressive prediction")
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, batch_size=32)
first_batch = next(iter(loader))
batch_X, batch_y = first_batch
first_batch = torch.from_numpy(batch_X)
logging.debug("first_batch {}".format(first_batch.size()))
use_cuda = torch.cuda.is_available()
model1 = t.keypoint_regression_model()
if use_cuda:
first_batch = first_batch.cuda()
model1 = model1.cuda()
y = model1(first_batch)
logging.debug("y shape {}".format(y.size()))
logging.debug("target y shape {}".format(batch_y.shape))
def main():
args = setup_args()
print(red("args {}".format(args)))
model = load_model(args.model_path)
'''
get a random test picture and imshow the predicted kps
'''
data_dir = "data"
loader = ds.kaggle_face_dataset(data_dir, 32, test=True)
testing_samples, labels, testing_ori, labels_ori = \
loader.X, loader.y, loader.X_ori, loader.y_ori
n_samples = len(testing_samples)
sampled_ind = random.randint(0, n_samples)
selected_pic = loader.X[sampled_ind]
# ds.plot_testing_sample(selected_pic)
print("selected pic shape {}".format(selected_pic.shape))
input_tensor = t.mk_cuda(torch.from_numpy(selected_pic.reshape((1, 1, 96, 96))))
predicted_y = model(input_tensor)
predicted_y = predicted_y.cpu().detach().numpy().reshape(30)
print("predicted {}".format(predicted_y))
ds.plot_sample_denormed(selected_pic, predicted_y)
def main():
cf = get_conf()
loader = ds.kaggle_face_dataset(cf.dataset.dir, cf.dataset.batch_size)
model = get_model(cf)
train_loop(loader, model, cf)