def test_image_io_vector(resx, resy, comp, ext, dt):
shape = (resx, resy)
pixel = np.random.rand(*shape, comp).astype(to_numpy_type(dt))
pixel_t = ti.Vector.field(comp, dt, shape)
pixel_t.from_numpy(pixel)
fn = make_temp_file(suffix='.' + ext)
ti.imwrite(pixel_t, fn)
pixel_r = (ti.imread(fn).astype(to_numpy_type(dt)) + 0.5) / 256.0
assert np.allclose(pixel_r, pixel, atol=2e-2)
os.remove(fn)
def to_numpy(self):
import numpy as np
from taichi.lang.meta import tensor_to_ext_arr
arr = np.zeros(shape=self.shape, dtype=to_numpy_type(self.dtype))
tensor_to_ext_arr(self, arr)
ti.sync()
return arr
def make_constant_expr(val, dtype):
if isinstance(val, (float, np.floating)):
constant_dtype = impl.get_runtime(
).default_fp if dtype is None else dtype
if constant_dtype not in real_types:
raise TaichiTypeError(
'Floating-point literals must be annotated with a floating-point type. For type casting, use `ti.cast`.'
)
return Expr(_ti_core.make_const_expr_fp(constant_dtype, val))
if isinstance(val, (int, np.integer)):
constant_dtype = impl.get_runtime(
).default_ip if dtype is None else dtype
if constant_dtype not in integer_types:
raise TaichiTypeError(
'Integer literals must be annotated with a integer type. For type casting, use `ti.cast`.'
)
if _check_in_range(to_numpy_type(constant_dtype), val):
return Expr(
_ti_core.make_const_expr_int(
constant_dtype, _clamp_unsigned_to_range(np.int64, val)))
if dtype is None:
raise TaichiTypeError(
f'Integer literal {val} exceeded the range of default_ip: {impl.get_runtime().default_ip}, please specify the dtype via e.g. `ti.u64({val})` or set a different `default_ip` in `ti.init()`'
)
else:
raise TaichiTypeError(
f'Integer literal {val} exceeded the range of specified dtype: {dtype}'
)
raise TaichiTypeError(f'Invalid constant scalar data type: {type(val)}')
def to_numpy(self, keep_dims=False, as_vector=None, dtype=None):
"""Converts `self` to a numpy array.
Args:
keep_dims (bool, optional): Whether to keep the dimension after conversion.
When keep_dims=True, on an n-D matrix field, the numpy array always has n+2 dims, even for 1x1, 1xn, nx1 matrix fields.
When keep_dims=False, the resulting numpy array should skip the matrix dims with size 1.
For example, a 4x1 or 1x4 matrix field with 5x6x7 elements results in an array of shape 5x6x7x4.
as_vector (bool, deprecated): Whether to make the returned numpy array as a vector, i.e., with shape (n,) rather than (n, 1).
Note that this argument has been deprecated.
More discussion about `as_vector`: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858.
dtype (DataType, optional): The desired data type of returned numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
if as_vector is not None:
warning(
'v.to_numpy(as_vector=True) is deprecated, '
'please use v.to_numpy() directly instead',
DeprecationWarning,
stacklevel=3)
if dtype is None:
dtype = to_numpy_type(self.dtype)
as_vector = self.m == 1 and not keep_dims
shape_ext = (self.n, ) if as_vector else (self.n, self.m)
import numpy as np
arr = np.zeros(self.shape + shape_ext, dtype=dtype)
from taichi.lang.meta import matrix_to_ext_arr
matrix_to_ext_arr(self, arr, as_vector)
ti.sync()
return arr
def to_numpy(self, dtype=None):
if dtype is None:
dtype = to_numpy_type(self.dtype)
import numpy as np # pylint: disable=C0415
arr = np.zeros(shape=self.shape, dtype=dtype)
taichi.lang.meta.tensor_to_ext_arr(self, arr)
ti.sync()
return arr
def to_numpy(self, dtype=None):
if dtype is None:
dtype = to_numpy_type(self.dtype)
import numpy as np
arr = np.zeros(shape=self.shape, dtype=dtype)
from taichi.lang.meta import tensor_to_ext_arr
tensor_to_ext_arr(self, arr)
ti.sync()
return arr
def to_numpy(self, dtype=None):
if dtype is None:
dtype = to_numpy_type(self.dtype)
import numpy as np # pylint: disable=C0415
arr = np.zeros(shape=self.shape, dtype=dtype)
from taichi._kernels import tensor_to_ext_arr # pylint: disable=C0415
tensor_to_ext_arr(self, arr)
taichi.lang.runtime_ops.sync()
return arr
def make_constant_expr(val, dtype):
if isinstance(val, (int, np.integer)):
constant_dtype = pytaichi.default_ip if dtype is None else dtype
_check_in_range(to_numpy_type(constant_dtype), val)
return Expr(
_ti_core.make_const_expr_int(
constant_dtype, _clamp_unsigned_to_range(np.int64, val)))
if isinstance(val, (float, np.floating)):
constant_dtype = pytaichi.default_fp if dtype is None else dtype
return Expr(_ti_core.make_const_expr_fp(constant_dtype, val))
raise TaichiTypeError(f'Invalid constant scalar data type: {type(val)}')
def _ndarray_to_numpy(self):
"""Converts ndarray to a numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
arr = np.zeros(shape=self.arr.shape, dtype=to_numpy_type(self.dtype))
from taichi._kernels import ndarray_to_ext_arr # pylint: disable=C0415
ndarray_to_ext_arr(self, arr)
impl.get_runtime().sync()
return arr
def ndarray_to_numpy(self):
"""Converts ndarray to a numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
if impl.current_cfg().ndarray_use_torch:
return self.arr.cpu().numpy()
arr = np.zeros(shape=self.arr.shape, dtype=to_numpy_type(self.dtype))
taichi.lang.meta.ndarray_to_ext_arr(self, arr)
impl.get_runtime().sync()
return arr
def _ndarray_matrix_to_numpy(self, layout, as_vector):
"""Converts matrix ndarray to a numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
arr = np.zeros(shape=self.arr.shape, dtype=to_numpy_type(self.dtype))
from taichi._kernels import \
ndarray_matrix_to_ext_arr # pylint: disable=C0415
layout_is_aos = 1 if layout == Layout.AOS else 0
ndarray_matrix_to_ext_arr(self, arr, layout_is_aos, as_vector)
impl.get_runtime().sync()
return arr
def to_numpy(self):
"""Create a numpy array containing the same elements when the class itself represents GlobalVariableExpression (field) or ExternalTensorExpression internally.
This is an unified interface to match :func:`taichi.lang.Matrix.to_numpy`.
Returns:
The numpy array containing the same elements when the class itself represents GlobalVariableExpression (field) or ExternalTensorExpression internally.
"""
import numpy as np
from taichi.lang.meta import tensor_to_ext_arr
arr = np.zeros(shape=self.shape, dtype=to_numpy_type(self.dtype))
tensor_to_ext_arr(self, arr)
ti.sync()
return arr
def test_image_io_uint(resx, resy, comp, ext, dt):
shape = (resx, resy)
np_type = to_numpy_type(dt)
# When saving to disk, pixel data will be truncated into 8 bits.
# Be careful here if you want lossless saving.
np_max = np.iinfo(np_type).max // 256
pixel = np.random.randint(256, size=(*shape, comp), dtype=np_type) * np_max
pixel_t = ti.Vector.field(comp, dt, shape)
pixel_t.from_numpy(pixel)
fn = make_temp_file(suffix='.' + ext)
ti.imwrite(pixel_t, fn)
pixel_r = ti.imread(fn).astype(np_type) * np_max
assert (pixel_r == pixel).all()
os.remove(fn)
def ndarray_matrix_to_numpy(self, as_vector):
"""Converts matrix ndarray to a numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
if self.ndarray_use_torch:
return self.arr.cpu().numpy()
arr = np.zeros(shape=self.arr.shape, dtype=to_numpy_type(self.dtype))
from taichi._kernels import \
ndarray_matrix_to_ext_arr # pylint: disable=C0415
ndarray_matrix_to_ext_arr(self, arr, as_vector)
impl.get_runtime().sync()
return arr
def to_numpy(self):
"""Converts ndarray to a numpy array.
Returns:
numpy.ndarray: The result numpy array.
"""
if impl.current_cfg().ndarray_use_torch:
return self.arr.cpu().numpy()
else:
import numpy as np # pylint: disable=C0415
arr = np.zeros(shape=self.arr.shape,
dtype=to_numpy_type(self.dtype))
from taichi.lang.meta import \
ndarray_to_ext_arr # pylint: disable=C0415
ndarray_to_ext_arr(self, arr)
impl.get_runtime().sync()
return arr
def to_numpy(self, keep_dims=False, as_vector=None, dtype=None):
# Discussion: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858
if as_vector is not None:
warning(
'v.to_numpy(as_vector=True) is deprecated, '
'please use v.to_numpy() directly instead',
DeprecationWarning,
stacklevel=3)
as_vector = self.m == 1 and not keep_dims
shape_ext = (self.n, ) if as_vector else (self.n, self.m)
if not self.is_global():
return np.array(self.entries).reshape(shape_ext)
if dtype is None:
dtype = to_numpy_type(self.dtype)
ret = np.zeros(self.shape + shape_ext, dtype=dtype)
from taichi.lang.meta import matrix_to_ext_arr
matrix_to_ext_arr(self, ret, as_vector)
return ret
def make_constant_expr(val, dtype):
if isinstance(val, (int, np.integer)):
constant_dtype = impl.get_runtime(
).default_ip if dtype is None else dtype
if constant_dtype not in integer_types:
raise TaichiTypeError(
'Integer literals must be annotated with a integer type. For type casting, use `ti.cast`.'
)
_check_in_range(to_numpy_type(constant_dtype), val)
return Expr(
_ti_core.make_const_expr_int(
constant_dtype, _clamp_unsigned_to_range(np.int64, val)))
if isinstance(val, (float, np.floating)):
constant_dtype = impl.get_runtime(
).default_fp if dtype is None else dtype
if constant_dtype not in real_types:
raise TaichiTypeError(
'Floating-point literals must be annotated with a floating-point type. For type casting, use `ti.cast`.'
)
return Expr(_ti_core.make_const_expr_fp(constant_dtype, val))
raise TaichiTypeError(f'Invalid constant scalar data type: {type(val)}')
def test_image_io(resx, resy, comp, ext, is_field, dt):
if comp != 1:
shape = (resx, resy, comp)
else:
shape = (resx, resy)
if is_field:
pixel_t = ti.field(dt, shape)
pixel = np.random.randint(256, size=shape, dtype=to_numpy_type(dt))
if is_field:
pixel_t.from_numpy(pixel)
fn = make_temp_file(suffix='.' + ext)
if is_field:
ti.imwrite(pixel_t, fn)
else:
ti.imwrite(pixel, fn)
pixel_r = ti.imread(fn)
if comp == 1:
# from (resx, resy, 1) to (resx, resy)
pixel_r = pixel_r.reshape((resx, resy))
assert (pixel_r == pixel).all()
os.remove(fn)
def to_numpy(self, keep_dims=False, as_vector=None, dtype=None):
"""Convert the taichi matrix to a numpy.ndarray.
Args:
keep_dims (bool, optional): Whether keep the dimension after conversion.
When keep_dims=True, on an n-D matrix field, the numpy array always has n+2 dims, even for 1x1, 1xn, nx1 matrix fields.
When keep_dims=False, the resulting numpy array should skip the dimensionality with only 1 element, on the matrix shape dimensionalities.
For example, a 4x1 or 1x4 matrix field with 5x6x7 elements results in an array of shape 5x6x7x4.
as_vector (bool, deprecated): Make the returned numpy array as a vector i.e., has a shape (n,) rather than (n, 1)
Note that this argument has been deprecated.
More discussion about `as_vector`: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858.
dtype (DataType, optional): The desired data type of returned numpy array.
Returns:
numpy.ndarray: The numpy array that converted from the matrix field.
"""
# Discussion: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858
if as_vector is not None:
warning(
'v.to_numpy(as_vector=True) is deprecated, '
'please use v.to_numpy() directly instead',
DeprecationWarning,
stacklevel=3)
as_vector = self.m == 1 and not keep_dims
shape_ext = (self.n, ) if as_vector else (self.n, self.m)
if not self.is_global():
return np.array(self.entries).reshape(shape_ext)
if dtype is None:
dtype = to_numpy_type(self.dtype)
ret = np.zeros(self.shape + shape_ext, dtype=dtype)
from taichi.lang.meta import matrix_to_ext_arr
matrix_to_ext_arr(self, ret, as_vector)
return ret
