• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
2 
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6 
7     http://www.apache.org/licenses/LICENSE-2.0
8 
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15 
16 #ifndef TENSORFLOW_CORE_KERNELS_COLORSPACE_OP_H_
17 #define TENSORFLOW_CORE_KERNELS_COLORSPACE_OP_H_
18 
19 #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
20 #include "tensorflow/core/framework/tensor_shape.h"
21 #include "tensorflow/core/framework/tensor_types.h"
22 
23 namespace tensorflow {
24 
25 namespace functor {
26 
27 template <typename Device, typename T>
28 struct RGBToHSV {
operatorRGBToHSV29   void operator()(const Device &d,
30                   typename TTypes<T, 2>::ConstTensor input_data,
31                   typename TTypes<T, 1>::Tensor range,
32                   typename TTypes<T, 2>::Tensor output_data) {
33     auto H = output_data.template chip<1>(0);
34     auto S = output_data.template chip<1>(1);
35     auto V = output_data.template chip<1>(2);
36 
37     auto R = input_data.template chip<1>(0);
38     auto G = input_data.template chip<1>(1);
39     auto B = input_data.template chip<1>(2);
40 
41 #if !defined(EIGEN_HAS_INDEX_LIST)
42     Eigen::array<int, 1> channel_axis{{1}};
43 #else
44     Eigen::IndexList<Eigen::type2index<1> > channel_axis;
45 #endif
46 
47     V.device(d) = input_data.maximum(channel_axis);
48 
49     range.device(d) = V - input_data.minimum(channel_axis);
50 
51     S.device(d) = (V > T(0)).select(range / V, V.constant(T(0)));
52 
53     auto norm = range.inverse() * (T(1) / T(6));
54     // TODO(wicke): all these assignments are only necessary because a combined
55     // expression is larger than kernel parameter space. A custom kernel is
56     // probably in order.
57     H.device(d) = (R == V).select(
58         norm * (G - B), (G == V).select(norm * (B - R) + T(2) / T(6),
59                                         norm * (R - G) + T(4) / T(6)));
60     H.device(d) = (range > T(0)).select(H, H.constant(T(0)));
61     H.device(d) = (H < T(0)).select(H + T(1), H);
62   }
63 };
64 
65 template <typename Device, typename T>
66 struct HSVToRGB {
operatorHSVToRGB67   void operator()(const Device &d,
68                   typename TTypes<T, 2>::ConstTensor input_data,
69                   typename TTypes<T, 2>::Tensor output_data) {
70     auto H = input_data.template chip<1>(0);
71     auto S = input_data.template chip<1>(1);
72     auto V = input_data.template chip<1>(2);
73 
74     // TODO(wicke): compute only the fractional part of H for robustness
75     auto dh = H * T(6);
76     auto dr = ((dh - T(3)).abs() - T(1)).cwiseMax(T(0)).cwiseMin(T(1));
77     auto dg = (-(dh - T(2)).abs() + T(2)).cwiseMax(T(0)).cwiseMin(T(1));
78     auto db = (-(dh - T(4)).abs() + T(2)).cwiseMax(T(0)).cwiseMin(T(1));
79     auto one_s = -S + T(1);
80 
81     auto R = output_data.template chip<1>(0);
82     auto G = output_data.template chip<1>(1);
83     auto B = output_data.template chip<1>(2);
84 
85     R.device(d) = (one_s + S * dr) * V;
86     G.device(d) = (one_s + S * dg) * V;
87     B.device(d) = (one_s + S * db) * V;
88   }
89 };
90 
91 }  // namespace functor
92 }  // namespace tensorflow
93 
94 #endif  // TENSORFLOW_CORE_KERNELS_COLORSPACE_OP_H_
95