def visualizeAnomalyImage(img, img_, z, z_):
"""
Visualize the testing result and render the level of abnormality
The anamoly score map will are also concatenated in the last
Arg: img (torch.Tensor) - The input image
img_ (torch.Tensor) - The reconstructed image
z (torch.Tensor) - The latend representation (tensor shape)
z_ (torch.Tensor) - The reconstructed latend representation (tensor shape)
"""
s_map = visualizeAnomalyMap(img, z, z_)
# Form the shown image
img = sunnerTransforms.asImg(img)[0]
img_ = sunnerTransforms.asImg(img_)[0]
result = np.hstack((img, img_, s_map * 255.0, img * s_map))
result = result.astype(np.uint8)
# --- Plot with matplotlib
plt.imshow(img, interpolation='nearest')
plt.imshow(s_map[:, :, 0],
cmap=plt.cm.viridis,
alpha=0.5,
interpolation='nearest')
plt.show()
def main():
# Create the fundemental data loader
loader = sunnerData.DataLoader(sunnerData.ImageDataset(
root=[['/home/sunner/Music/waiting_for_you_dataset/wait'],
['/home/sunner/Music/waiting_for_you_dataset/real_world']],
transforms=transforms.Compose([
sunnertransforms.Resize((160, 320)),
sunnertransforms.ToTensor(),
sunnertransforms.ToFloat(),
sunnertransforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5]),
])),
batch_size=32,
shuffle=False,
num_workers=2)
# Use upper wrapper to assign particular iteration
loader = sunnerData.IterationLoader(loader, max_iter=1)
# Show!
for batch_tensor, _ in loader:
batch_img = sunnertransforms.asImg(batch_tensor, size=(160, 320))
cv2.imshow('show_window', batch_img[0][:, :, ::-1])
cv2.waitKey(0)
# Or show multiple image in one line
sunnertransforms.show(batch_tensor[:10], row=2, column=5)
def visualizeEncoderDecoder(img, img_, z, z_):
"""
Visualize the rendered result after adversarial auto-encoder
The anamoly score map will are also concatenated in the last
Arg: img (torch.Tensor) - The input image
img_ (torch.Tensor) - The reconstructed image
z (torch.Tensor) - The latend representation (tensor shape)
z_ (torch.Tensor) - The reconstructed latend representation (tensor shape)
"""
s_map = visualizeAnomalyMap(img, z, z_)
img = sunnerTransforms.asImg(img)[0, :, :, ::-1]
img_ = sunnerTransforms.asImg(img_)[0, :, :, ::-1]
result = np.hstack((img, img_, s_map * 255.0))
result = result.astype(np.uint8)
cv2.imshow('training visualization', result)
cv2.waitKey(10)
def main():
# Define the loader to generate the pallete object
loader = sunnerData.DataLoader(
sunnerData.ImageDataset(
root=[tag_folder],
transform=transforms.Compose([
sunnertransforms.ToTensor(),
]),
save_file=False # Don't save the record file, be careful!
),
batch_size=2,
shuffle=False,
num_workers=2)
pallete = sunnertransforms.getCategoricalMapping(loader,
path='pallete.json')[0]
del loader
# Define the actual loader
loader = sunnerData.DataLoader(sunnerData.ImageDataset(
root=[img_folder, tag_folder],
transform=transforms.Compose([
sunnertransforms.Resize((512, 1024)),
sunnertransforms.ToTensor(),
sunnertransforms.Transpose(sunnertransforms.BHWC2BCHW),
sunnertransforms.Normalize(),
])),
batch_size=32,
shuffle=False,
num_workers=2)
# Define the reverse operator
goto_op = sunnertransforms.CategoricalTranspose(
pallete=pallete, direction=sunnertransforms.COLOR2ONEHOT)
back_op = sunnertransforms.CategoricalTranspose(
pallete=pallete, direction=sunnertransforms.ONEHOT2COLOR)
# Show!
for _, batch_index in loader:
batch_img = back_op(goto_op(batch_index))
batch_img = sunnertransforms.asImg(batch_img, size=(512, 1024))
cv2.imshow('show_window', batch_img[0][:, :, ::-1])
cv2.waitKey(0)
break
def main():
# Define the loader to generate the pallete object
loader = sunnerData.DataLoader(sunnerData.ImageDataset(
root = [
tag_folder
],
transforms = transforms.Compose([
sunnertransforms.ToTensor(),
sunnertransforms.ToFloat(),
sunnertransforms.UnNormalize(mean=[0, 0, 0], std=[255, 255, 255]), # Remember to transfer back to [0~255] before generate pallete
sunnertransforms.Transpose(sunnertransforms.BCHW2BHWC) # Remember to transfer back to BHWC before generate pallete
])
), batch_size = 2, shuffle = False, num_workers = 2
)
pallete = sunnertransforms.getCategoricalMapping(loader, path = 'pallete.json')[0]
del loader
# Define the actual loader
loader = sunnerData.DataLoader(sunnerData.ImageDataset(
root = [
img_folder,
tag_folder
],
transforms = transforms.Compose([
sunnertransforms.Resize((512, 1024)),
sunnertransforms.ToTensor(),
sunnertransforms.ToFloat(),
sunnertransforms.Normalize(mean = [0.5, 0.5, 0.5], std = [0.5, 0.5, 0.5]),
])), batch_size = 32, shuffle = False, num_workers = 2
)
# Define the reverse operator
goto_op = sunnertransforms.CategoricalTranspose(pallete = pallete, direction = sunnertransforms.COLOR2ONEHOT)
back_op = sunnertransforms.CategoricalTranspose(pallete = pallete, direction = sunnertransforms.ONEHOT2COLOR)
# Show!
for _, batch_index in loader:
batch_img = back_op(goto_op(batch_index))
batch_img = sunnertransforms.asImg(batch_img, size = (512, 1024))
cv2.imshow('show_window', batch_img[0][:, :, ::-1])
cv2.waitKey(0)
break