diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 979802f98..ad0eb8504 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -119,6 +119,8 @@ ncnn_add_layer(Threshold)
 ncnn_add_layer(Tile OFF)
 ncnn_add_layer(RNN OFF)
 ncnn_add_layer(LSTM OFF)
+ncnn_add_layer(BinaryOp)
+ncnn_add_layer(UnaryOp)
 
 add_library(ncnn STATIC ${ncnn_SRCS})
 
diff --git a/src/layer.h b/src/layer.h
index 13bb2e552..ca3e1298d 100644
--- a/src/layer.h
+++ b/src/layer.h
@@ -130,6 +130,8 @@ enum
     Tile        = 37,
     RNN         = 38,
     LSTM        = 39,
+    BinaryOp    = 40,
+    UnaryOp     = 41,
 
     CustomBit   = (1<<8),
 };
diff --git a/src/layer/binaryop.cpp b/src/layer/binaryop.cpp
new file mode 100644
index 000000000..36f006fb2
--- /dev/null
+++ b/src/layer/binaryop.cpp
@@ -0,0 +1,182 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#include "binaryop.h"
+#include <math.h>
+
+namespace ncnn {
+
+DEFINE_LAYER_CREATOR(BinaryOp)
+
+BinaryOp::BinaryOp()
+{
+    one_blob_only = false;
+    support_inplace = false;
+}
+
+#if NCNN_STDIO
+#if NCNN_STRING
+int BinaryOp::load_param(FILE* paramfp)
+{
+    int nscan = fscanf(paramfp, "%d", &op_type);
+    if (nscan != 1)
+    {
+        fprintf(stderr, "BinaryOp load_param failed %d\n", nscan);
+        return -1;
+    }
+
+    return 0;
+}
+#endif // NCNN_STRING
+int BinaryOp::load_param_bin(FILE* paramfp)
+{
+    fread(&op_type, sizeof(int), 1, paramfp);
+
+    return 0;
+}
+#endif // NCNN_STDIO
+
+int BinaryOp::load_param(const unsigned char*& mem)
+{
+    op_type = *(int*)(mem);
+    mem += 4;
+
+    return 0;
+}
+
+int BinaryOp::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs) const
+{
+    const Mat& bottom_blob = bottom_blobs[0];
+    const Mat& bottom_blob1 = bottom_blobs[1];
+
+    int w = bottom_blob.w;
+    int h = bottom_blob.h;
+    int channels = bottom_blob.c;
+    int size = w * h;
+
+    Mat& top_blob = top_blobs[0];
+    top_blob.create(w, h, channels);
+    if (top_blob.empty())
+        return -100;
+
+    if (op_type == Operation_ADD)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ptr[i] + ptr1[i];
+            }
+        }
+    }
+    else if (op_type == Operation_SUB)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ptr[i] - ptr1[i];
+            }
+        }
+    }
+    else if (op_type == Operation_MUL)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ptr[i] * ptr1[i];
+            }
+        }
+    }
+    else if (op_type == Operation_DIV)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ptr[i] / ptr1[i];
+            }
+        }
+    }
+    else if (op_type == Operation_MAX)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = std::max(ptr[i], ptr1[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_MIN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = std::min(ptr[i], ptr1[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_POW)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            const float* ptr1 = bottom_blob1.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = pow(ptr[i], ptr1[i]);
+            }
+        }
+    }
+
+    return 0;
+}
+
+} // namespace ncnn
diff --git a/src/layer/binaryop.h b/src/layer/binaryop.h
new file mode 100644
index 000000000..a99b8c43f
--- /dev/null
+++ b/src/layer/binaryop.h
@@ -0,0 +1,54 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#ifndef LAYER_BINARYOP_H
+#define LAYER_BINARYOP_H
+
+#include "layer.h"
+
+namespace ncnn {
+
+class BinaryOp : public Layer
+{
+public:
+    BinaryOp();
+
+#if NCNN_STDIO
+#if NCNN_STRING
+    virtual int load_param(FILE* paramfp);
+#endif // NCNN_STRING
+    virtual int load_param_bin(FILE* paramfp);
+#endif // NCNN_STDIO
+    virtual int load_param(const unsigned char*& mem);
+
+    virtual int forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs) const;
+
+    enum {
+        Operation_ADD   = 0,
+        Operation_SUB   = 1,
+        Operation_MUL   = 2,
+        Operation_DIV   = 3,
+        Operation_MAX   = 4,
+        Operation_MIN   = 5,
+        Operation_POW   = 6
+    };
+
+public:
+    // param
+    int op_type;
+};
+
+} // namespace ncnn
+
+#endif // LAYER_BINARYOP_H
diff --git a/src/layer/unaryop.cpp b/src/layer/unaryop.cpp
new file mode 100644
index 000000000..4fbcca737
--- /dev/null
+++ b/src/layer/unaryop.cpp
@@ -0,0 +1,489 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#include "unaryop.h"
+#include <math.h>
+
+namespace ncnn {
+
+DEFINE_LAYER_CREATOR(UnaryOp)
+
+UnaryOp::UnaryOp()
+{
+    one_blob_only = true;
+    support_inplace = true;
+}
+
+#if NCNN_STDIO
+#if NCNN_STRING
+int UnaryOp::load_param(FILE* paramfp)
+{
+    int nscan = fscanf(paramfp, "%d", &op_type);
+    if (nscan != 1)
+    {
+        fprintf(stderr, "UnaryOp load_param failed %d\n", nscan);
+        return -1;
+    }
+
+    return 0;
+}
+#endif // NCNN_STRING
+int UnaryOp::load_param_bin(FILE* paramfp)
+{
+    fread(&op_type, sizeof(int), 1, paramfp);
+
+    return 0;
+}
+#endif // NCNN_STDIO
+
+int UnaryOp::load_param(const unsigned char*& mem)
+{
+    op_type = *(int*)(mem);
+    mem += 4;
+
+    return 0;
+}
+
+int UnaryOp::forward(const Mat& bottom_blob, Mat& top_blob) const
+{
+    int w = bottom_blob.w;
+    int h = bottom_blob.h;
+    int channels = bottom_blob.c;
+    int size = w * h;
+
+    top_blob.create(w, h, channels);
+    if (top_blob.empty())
+        return -100;
+
+    if (op_type == Operation_ABS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = fabs(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_NEG)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = -ptr[i];
+            }
+        }
+    }
+    else if (op_type == Operation_FLOOR)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = floor(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_CEIL)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ceil(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_SQUARE)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = ptr[i] * ptr[i];
+            }
+        }
+    }
+    else if (op_type == Operation_SQRT)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = sqrt(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_RSQRT)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = 1.f / sqrt(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_EXP)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = exp(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_LOG)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = log(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_SIN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = sin(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_COS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = cos(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_TAN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = tan(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ASIN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = asin(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ACOS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = acos(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ATAN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            const float* ptr = bottom_blob.channel(q);
+            float* outptr = top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                outptr[i] = atan(ptr[i]);
+            }
+        }
+    }
+
+    return 0;
+}
+
+int UnaryOp::forward_inplace(Mat& bottom_top_blob) const
+{
+    int w = bottom_top_blob.w;
+    int h = bottom_top_blob.h;
+    int channels = bottom_top_blob.c;
+    int size = w * h;
+
+    if (op_type == Operation_ABS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = fabs(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_NEG)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = -ptr[i];
+            }
+        }
+    }
+    else if (op_type == Operation_FLOOR)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = floor(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_CEIL)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = ceil(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_SQUARE)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = ptr[i] * ptr[i];
+            }
+        }
+    }
+    else if (op_type == Operation_SQRT)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = sqrt(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_RSQRT)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = 1.f / sqrt(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_EXP)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = exp(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_LOG)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = log(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_SIN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = sin(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_COS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = cos(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_TAN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = tan(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ASIN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = asin(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ACOS)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = acos(ptr[i]);
+            }
+        }
+    }
+    else if (op_type == Operation_ATAN)
+    {
+        #pragma omp parallel for
+        for (int q=0; q<channels; q++)
+        {
+            float* ptr = bottom_top_blob.channel(q);
+
+            for (int i=0; i<size; i++)
+            {
+                ptr[i] = atan(ptr[i]);
+            }
+        }
+    }
+
+    return 0;
+}
+
+} // namespace ncnn
diff --git a/src/layer/unaryop.h b/src/layer/unaryop.h
new file mode 100644
index 000000000..76dd7bc27
--- /dev/null
+++ b/src/layer/unaryop.h
@@ -0,0 +1,64 @@
+// Tencent is pleased to support the open source community by making ncnn available.
+//
+// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
+//
+// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
+// in compliance with the License. You may obtain a copy of the License at
+//
+// https://opensource.org/licenses/BSD-3-Clause
+//
+// 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.
+
+#ifndef LAYER_UNARYOP_H
+#define LAYER_UNARYOP_H
+
+#include "layer.h"
+
+namespace ncnn {
+
+class UnaryOp : public Layer
+{
+public:
+    UnaryOp();
+
+#if NCNN_STDIO
+#if NCNN_STRING
+    virtual int load_param(FILE* paramfp);
+#endif // NCNN_STRING
+    virtual int load_param_bin(FILE* paramfp);
+#endif // NCNN_STDIO
+    virtual int load_param(const unsigned char*& mem);
+
+    virtual int forward(const Mat& bottom_blob, Mat& top_blob) const;
+
+    virtual int forward_inplace(Mat& bottom_top_blob) const;
+
+    enum {
+        Operation_ABS   = 0,
+        Operation_NEG   = 1,
+        Operation_FLOOR = 2,
+        Operation_CEIL  = 3,
+        Operation_SQUARE= 4,
+        Operation_SQRT  = 5,
+        Operation_RSQRT = 6,
+        Operation_EXP   = 7,
+        Operation_LOG   = 8,
+        Operation_SIN   = 9,
+        Operation_COS   = 10,
+        Operation_TAN   = 11,
+        Operation_ASIN  = 12,
+        Operation_ACOS  = 13,
+        Operation_ATAN  = 14
+    };
+
+public:
+    // param
+    int op_type;
+};
+
+} // namespace ncnn
+
+#endif // LAYER_UNARYOP_H