def postprocess(img: torch.tensor):
# img = img[0][2]
img = torch.argmax(img, dim=1)
img = img.cpu().numpy()
img = np.squeeze(img)
img = re_normalize(img)
# img = util.invert(img)
return img
def transform_x(self, x):
# make sure it's a numpy.ndarray on the cpu
x = x.cpu().numpy()
# from [C, H, W] to [H, W, C] - only for RGB images.
if self.check_if_rgb(x):
x = np.moveaxis(x, source=0, destination=-1)
# Re-normalize
x = re_normalize(x)
return x
def transform_image(self, x):
image_transformed = self.rcnn_transform([self.image])
x = image_transformed[0].tensors[0]
# make sure it's a numpy.ndarray on the cpu
x = x.cpu().numpy()
# from [C, H, W] to [H, W, C] - only for RGB images.
x = np.moveaxis(x, source=0, destination=-1)
# Re-normalize
x = re_normalize(x)
return x
def transform_x(self, sample):
# dict unpacking
x, x_name = sample['x'], sample['x_name']
# make sure it's a numpy.ndarray on the cpu
x = x.cpu().numpy()
# from [C, H, W] to [H, W, C] - only for RGB images.
if self.check_if_rgb(x):
x = np.moveaxis(x, source=0, destination=-1)
# Re-normalize
x = re_normalize(x)
return x, x_name
def __next__(self):
x, y = next(iter(self.dataloader))
x, y = x.cpu().numpy(), y.cpu().numpy() # make sure it's a numpy.ndarray on the cpu
# Batch
batch_size = x.shape[0]
rand_num = np.random.randint(low=0, high=batch_size)
x, y = x[rand_num], y[rand_num] # Pick a random image from the batch
# RGB
if self.rgb:
x = np.moveaxis(x, source=0, destination=-1) # from [C, H, W] to [H, W, C]
# Re-normalize
if self.re_normalize:
from transformations import re_normalize
x = re_normalize(x)
return x, y
def transform_x(self, x):
# Re-normalize
x = re_normalize(x)
return x
def postprocess(img: torch.tensor):
img = torch.argmax(img, dim=1) # perform argmax to generate 1 channel
img = img.cpu().numpy() # send to cpu and transform to numpy.ndarray
img = np.squeeze(img) # remove batch dim and channel dim -> [H, W]
img = re_normalize(img) # scale it to the range [0-255]
return img