importing kornia dos not compile

[BarSimanTov]

Describe the bug

importing kornia dos not compile.
It returns an error about "Unknown type name 'torch.device'"

torch in install and when i run "torch.device('cpu')" on my code - it works and return a valid result

What can it be?
I understand it didn't happened in previous versions. is that a way to install a private version on coda? (conda install -c conda-forge kornia==0.1.X failed)

Thanks!

Reproduction steps

1.import kornia as K 
2. run rour python file

Expected behavior

Traceback (most recent call last):
File "test.py", line 4, in
import kornia as K
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/kornia/init.py", line 13, in
from kornia import augmentation, color, contrib, enhance, feature, filters, geometry, jit, losses, morphology, utils
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/kornia/jit/init.py", line 11, in
spatial_soft_argmax2d = torch.jit.script(K.geometry.spatial_soft_argmax2d)
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/torch/jit/init.py", line 1226, in script
fn = torch._C._jit_script_compile(qualified_name, ast, _rcb, get_default_args(obj))
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/torch/jit/_recursive.py", line 222, in try_compile_fn
return torch.jit.script(fn, _rcb=rcb)
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/torch/jit/init.py", line 1226, in script
fn = torch._C._jit_script_compile(qualified_name, ast, _rcb, get_default_args(obj))
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/torch/jit/_recursive.py", line 222, in try_compile_fn
return torch.jit.script(fn, _rcb=rcb)
File "/home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/torch/jit/init.py", line 1226, in script
fn = torch._C._jit_script_compile(qualified_name, ast, _rcb, get_default_args(obj))
RuntimeError:
Unknown type name 'torch.device':
at /home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/kornia/utils/grid.py:10:21
def create_meshgrid(
height: int,
width: int,
normalized_coordinates: bool = True,
device: Optional[torch.device] = torch.device('cpu'),
~~~~~~~~~~~~ <--- HERE
dtype: torch.dtype = torch.float32,
) -> torch.Tensor:
"""Generates a coordinate grid for an image.

When the flag ``normalized_coordinates`` is set to True, the grid is
normalized to be in the range :math:`[-1,1]` to be consistent with the pytorch
function :py:func:`torch.nn.functional.grid_sample`.

Args:

'create_meshgrid' is being compiled since it was called from 'spatial_expectation2d'
at /home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/kornia/geometry/subpix/dsnt.py:82:4
... [0., 1., 0.]]]])
>>> spatial_expectation2d(heatmaps, False)
tensor([[[1., 2.]]])
"""
_validate_batched_image_tensor_input(input)

batch_size, channels, height, width = input.shape

# Create coordinates grid.
grid: torch.Tensor = create_meshgrid(height, width, normalized_coordinates, input.device)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
grid = grid.to(input.dtype)

pos_x: torch.Tensor = grid[..., 0].reshape(-1)
pos_y: torch.Tensor = grid[..., 1].reshape(-1)

input_flat: torch.Tensor = input.view(batch_size, channels, -1)

# Compute the expectation of the coordinates.
expected_y: torch.Tensor = torch.sum(pos_y * input_flat, -1, keepdim=True)

'spatial_expectation2d' is being compiled since it was called from 'spatial_soft_argmax2d'
at /home/barsiman/.conda/envs/puzzle2/lib/python3.7/site-packages/kornia/geometry/subpix/spatial_soft_argmax.py:529:4
Examples:
>>> input = torch.tensor([[[
... [0., 0., 0.],
... [0., 10., 0.],
... [0., 0., 0.]]]])
>>> spatial_soft_argmax2d(input, normalized_coordinates=False)
tensor([[[1.0000, 1.0000]]])
"""
input_soft: torch.Tensor = dsnt.spatial_softmax2d(input, temperature)
output: torch.Tensor = dsnt.spatial_expectation2d(input_soft, normalized_coordinates)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
return output

Environment

- PyTorch Version : 1.3.1
- OS (e.g., Linux): Linux
- How you installed PyTorch: conda install -c pytorch pytorch 
                        AND conda install -c anaconda pytorch-gpu

- Build command you used Not relevent 
- Python version: 3.7.3
- CUDA/cuDNN version: 4.8.3 ( conda-build version : 3.18.9)
- GPU models and configuration:Unknown 
- Any other relevant information:

Additional context

No response

[edgarriba]

Hi, from what I see you use a very old pytorch version. We give support to >1.6

Besides, the conda package is not maintained by us, so I would recommend using pip to install.

[BarSimanTov]

Hii

thanks for the quick replay!

for now, I need to stick with conda... so unforcedly pip is not really an option now
but upgrade torch seems to work! :-)

Thank you
Take care

[edgarriba]

Great !