From 3e00d8eb8254cad2ffa4cc2ec9175e36736d8a15 Mon Sep 17 00:00:00 2001 From: fengyihang Date: Tue, 22 Nov 2022 19:25:19 +0800 Subject: [PATCH] add bicubic in interpolate --- .../ops/mindspore.ops.func_interpolate.rst | 2 +- .../python/mindspore/ops/function/nn_func.py | 26 ++++---- tests/st/ops/test_func_interpolate_bicubic.py | 62 +++++++++++++++++++ 3 files changed, 78 insertions(+), 12 deletions(-) create mode 100644 tests/st/ops/test_func_interpolate_bicubic.py diff --git a/docs/api/api_python/ops/mindspore.ops.func_interpolate.rst b/docs/api/api_python/ops/mindspore.ops.func_interpolate.rst index a83573e8f0b..907172c9d3e 100644 --- a/docs/api/api_python/ops/mindspore.ops.func_interpolate.rst +++ b/docs/api/api_python/ops/mindspore.ops.func_interpolate.rst @@ -28,7 +28,7 @@ mindspore.ops.interpolate old_i = new_length != 0 ? new_i * old_length / new_length : 0 # if set to 'asymmetric' - - **mode** (str) - 所使用的插值方式。目前支持"linear"和"bilinear"插值方式。默认值："linear"。 + - **mode** (str) - 所使用的插值方式。目前支持"linear"，"bilinear"和"bicubic"插值方式。默认值："linear"。 返回： Tensor，数据类型与 `x` 相同。 diff --git a/mindspore/python/mindspore/ops/function/nn_func.py b/mindspore/python/mindspore/ops/function/nn_func.py index a1c40f62e6f..3cee27ed5fc 100644 --- a/mindspore/python/mindspore/ops/function/nn_func.py +++ b/mindspore/python/mindspore/ops/function/nn_func.py @@ -1903,7 +1903,7 @@ def _check_interpolate_inputs(input_dims, roi, scales, sizes, coordinate_transfo validator.check_value_type("mode", mode, [str], prim_name) if mode == "linear": validator.check_int(input_dims, 3, Rel.EQ, "input dims", prim_name) - elif mode == "bilinear": + elif mode in ["bilinear", "bicubic"]: validator.check_int(input_dims, 4, Rel.EQ, "input dims", prim_name) else: raise ValueError(f"{msg_prefix} mode must be 'linear' or 'bilinear', but got {mode}") @@ -1938,12 +1938,16 @@ def _interpolate_output_shape(shape, scales, sizes, mode): if sizes is not None: if mode == "bilinear": return sizes + if mode == "bicubic": + return Tensor(sizes, dtype=mstype.int32) return Tensor(sizes) ret = () for i in range(2, len(shape)): ret = ret + (int(scales[i] * shape[i]),) if mode == "bilinear": return ret + if mode == "bicubic": + return Tensor(ret, dtype=mstype.int32) return Tensor(ret) @@ -1986,7 +1990,7 @@ def interpolate(x, roi=None, scales=None, sizes=None, coordinate_transformation_ old_i = new_length != 0 ? new_i * old_length / new_length : 0 # if set to 'asymmetric' - mode (str): The method used to interpolate: 'linear' | 'bilinear'. Default is 'linear'. + mode (str): The method used to interpolate: 'linear' | 'bilinear' | 'bicubic'. Default is 'linear'. Returns: Resized tensor, with the same data type as input `x`. @@ -2027,26 +2031,26 @@ def interpolate(x, roi=None, scales=None, sizes=None, coordinate_transformation_ raise TypeError("For interpolate, the input x must be tensor") input_shape = x.shape input_dims = len(input_shape) - _check_interpolate_inputs(input_dims, roi, scales, sizes, coordinate_transformation_mode, mode, - "interpolate") + _check_interpolate_inputs(input_dims, roi, scales, sizes, coordinate_transformation_mode, mode, "interpolate") output_size = _interpolate_output_shape(input_shape, scales, sizes, mode) - if mode == "linear": resize_linear_inner = _get_cache_prim(IMG.ResizeLinear1D)( coordinate_transformation_mode=coordinate_transformation_mode) return resize_linear_inner(x, output_size) - if mode == "bilinear": + if mode in ["bilinear", "bicubic"]: align_corners = False half_pixel_centers = False if coordinate_transformation_mode == "align_corners": align_corners = True elif coordinate_transformation_mode == "half_pixel": half_pixel_centers = True - resize_bilinear_inner = _get_cache_prim(IMG.ResizeBilinearV2)(align_corners, half_pixel_centers) - return resize_bilinear_inner(x, output_size) - - raise TypeError( - "Input Error: For interpolate, {} mode is not support now".format(mode)) + if mode == "bilinear": + resize_bilinear_inner = _get_cache_prim(IMG.ResizeBilinearV2)(align_corners, half_pixel_centers) + return resize_bilinear_inner(x, output_size) + if mode == "bicubic": + resize_bicubic_inner = _get_cache_prim(IMG.ResizeBicubic)(align_corners, half_pixel_centers) + return resize_bicubic_inner(x.transpose((0, 2, 3, 1)), output_size).transpose((0, 3, 1, 2)) + raise TypeError("Input Error: For interpolate, {} mode is not support now".format(mode)) def softsign(x): diff --git a/tests/st/ops/test_func_interpolate_bicubic.py b/tests/st/ops/test_func_interpolate_bicubic.py new file mode 100644 index 00000000000..1311b0de26f --- /dev/null +++ b/tests/st/ops/test_func_interpolate_bicubic.py @@ -0,0 +1,62 @@ +# Copyright 2022 Huawei Technologies Co., Ltd +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# ============================================================================ + +import numpy as np +import pytest + +import mindspore as ms +import mindspore.nn as nn +import mindspore.ops as ops + + +class Net(nn.Cell): + def __init__(self, coordinate_transformation_mode, mode): + super(Net, self).__init__() + self.coordinate_transformation_mode = coordinate_transformation_mode + self.mode = mode + + def construct(self, x, scales=None, sizes=None): + return ops.interpolate(x, None, scales, sizes, self.coordinate_transformation_mode, self.mode) + + +@pytest.mark.level0 +@pytest.mark.platform_x86_cpu +@pytest.mark.platform_arm_cpu +@pytest.mark.platform_x86_gpu_training +@pytest.mark.env_onecard +@pytest.mark.parametrize('mode', [ms.GRAPH_MODE, ms.PYNATIVE_MODE]) +def test_interpolate_normal(mode): + """ + Feature: interpolate + Description: Verify the result of interpolate + Expectation: success + """ + ms.set_context(mode=mode) + net = Net("align_corners", "bicubic") + x = ops.arange(4, dtype=ms.float32).reshape((1, 1, 2, 2)) + out1 = net(x, sizes=(4, 6)) + expect_out1 = np.array([[[[0.0000, 0.1762, 0.3884, 0.6116, 0.8238, 1.0000], + [0.6289, 0.8051, 1.0174, 1.2405, 1.4527, 1.6289], + [1.3711, 1.5473, 1.7595, 1.9826, 2.1949, 2.3711], + [2.0000, 2.1762, 2.3884, 2.6116, 2.8238, 3.0000]]]]) + assert np.allclose(out1.asnumpy(), expect_out1, rtol=1e-4) + out2 = net(x, scales=(1.0, 1.0, 3.0, 3.0)) + expect_out2 = np.array([[[[0.0000, 0.1762, 0.3884, 0.6116, 0.8238, 1.0000], + [0.3524, 0.5286, 0.7408, 0.9640, 1.1762, 1.3524], + [0.7769, 0.9531, 1.1653, 1.3884, 1.6007, 1.7769], + [1.2231, 1.3993, 1.6116, 1.8347, 2.0469, 2.2231], + [1.6476, 1.8238, 2.0360, 2.2592, 2.4714, 2.6476], + [2.0000, 2.1762, 2.3884, 2.6116, 2.8238, 3.0000]]]]) + assert np.allclose(out2.asnumpy(), expect_out2, rtol=1e-4)