dnn_backend_native_layer_mathunary: add abs support

more math unary operations will be added here

It can be tested with the model file generated with below python scripy:

import tensorflow as tf
import numpy as np
import imageio

in_img = imageio.imread('input.jpeg')
in_img = in_img.astype(np.float32)/255.0
in_data = in_img[np.newaxis, :]

x = tf.placeholder(tf.float32, shape=[1, None, None, 3], name='dnn_in')
x1 = tf.subtract(x, 0.5)
x2 = tf.abs(x1)
y = tf.identity(x2, name='dnn_out')

sess=tf.Session()
sess.run(tf.global_variables_initializer())

graph_def = tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['dnn_out'])
tf.train.write_graph(graph_def, '.', 'image_process.pb', as_text=False)

print("image_process.pb generated, please use \
path_to_ffmpeg/tools/python/convert.py to generate image_process.model\n")

output = sess.run(y, feed_dict={x: in_data})
imageio.imsave("out.jpg", np.squeeze(output))

Signed-off-by: Ting Fu <ting.fu@intel.com>
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>

libavfilter/dnn/Makefile

@@ -6,6 +6,7 @@ OBJS-$(CONFIG_DNN)                           += dnn/dnn_backend_native_layer_con
 OBJS-$(CONFIG_DNN)                           += dnn/dnn_backend_native_layer_depth2space.o
 OBJS-$(CONFIG_DNN)                           += dnn/dnn_backend_native_layer_maximum.o
 OBJS-$(CONFIG_DNN)                           += dnn/dnn_backend_native_layer_mathbinary.o
+OBJS-$(CONFIG_DNN)                           += dnn/dnn_backend_native_layer_mathunary.o
 
 DNN-OBJS-$(CONFIG_LIBTENSORFLOW)             += dnn/dnn_backend_tf.o
 

libavfilter/dnn/dnn_backend_native.h

@@ -42,6 +42,7 @@ typedef enum {
     DLT_MIRROR_PAD = 3,
     DLT_MAXIMUM = 4,
     DLT_MATH_BINARY = 5,
+    DLT_MATH_UNARY = 6,
     DLT_COUNT
 } DNNLayerType;
 

libavfilter/dnn/dnn_backend_native_layer_mathunary.c

@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2020
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * DNN native backend implementation.
+ */
+
+#include "dnn_backend_native.h"
+#include "libavutil/avassert.h"
+#include "dnn_backend_native_layer_mathunary.h"
+
+int dnn_load_layer_math_unary(Layer *layer, AVIOContext *model_file_context, int file_size)
+{
+    DnnLayerMathUnaryParams *params;
+    int dnn_size = 0;
+    params = av_malloc(sizeof(*params));
+    if(!params)
+        return 0;
+
+    params->un_op = (int32_t)avio_rl32(model_file_context);
+    dnn_size += 4;
+    layer->params = params;
+    layer->input_operand_indexes[0] = (int32_t)avio_rl32(model_file_context);
+    layer->output_operand_index = (int32_t)avio_rl32(model_file_context);
+    dnn_size += 8;
+
+    return dnn_size;
+
+}
+
+int dnn_execute_layer_math_unary(DnnOperand *operands, const int32_t *input_operand_indexes,
+                                int32_t output_operand_index, const void *parameters)
+{
+    const DnnOperand *input = &operands[input_operand_indexes[0]];
+    DnnOperand *output = &operands[output_operand_index];
+    const DnnLayerMathUnaryParams *params = (const DnnLayerMathUnaryParams *)parameters;
+    int dims_count;
+    const float *src;
+    float *dst;
+
+    for (int i = 0; i < 4; ++i)
+        output->dims[i] = input->dims[i];
+
+    output->data_type = input->data_type;
+    output->length = calculate_operand_data_length(output);
+    output->data = av_realloc(output->data, output->length);
+    if (!output->data)
+        return DNN_ERROR;
+
+    dims_count = calculate_operand_dims_count(output);
+    src = input->data;
+    dst = output->data;
+
+    switch (params->un_op) {
+    case DMUO_ABS:
+        for (int i = 0; i < dims_count; ++i)
+            dst[i] = FFABS(src[i]);
+        return 0;
+    default:
+        return -1;
+    }
+}

libavfilter/dnn/dnn_backend_native_layer_mathunary.h

@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2020
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @file
+ * DNN inference functions interface for native backend.
+ */
+
+#ifndef AVFILTER_DNN_DNN_BACKEND_NATIVE_LAYER_MATHUNARY_H
+#define AVFILTER_DNN_DNN_BACKEND_NATIVE_LAYER_MATHUNARY_H
+
+#include "libavformat/avio.h"
+#include "dnn_backend_native.h"
+
+typedef enum {
+    DMUO_ABS = 0,
+    DMUO_COUNT
+} DNNMathUnaryOperation;
+
+typedef struct DnnLayerMathUnaryParams{
+    DNNMathUnaryOperation un_op;
+} DnnLayerMathUnaryParams;
+
+int dnn_load_layer_math_unary(Layer *layer, AVIOContext *model_file_context, int file_size);
+int dnn_execute_layer_math_unary(DnnOperand *operands, const int32_t *input_operand_indexes,
+                                int32_t output_operand_index, const void *parameters);
+
+#endif

libavfilter/dnn/dnn_backend_native_layers.c

@@ -25,6 +25,7 @@
 #include "dnn_backend_native_layer_depth2space.h"
 #include "dnn_backend_native_layer_maximum.h"
 #include "dnn_backend_native_layer_mathbinary.h"
+#include "dnn_backend_native_layer_mathunary.h"
 
 LayerFunc layer_funcs[DLT_COUNT] = {
     {NULL, NULL},
@@ -33,4 +34,5 @@ LayerFunc layer_funcs[DLT_COUNT] = {
     {dnn_execute_layer_pad,         dnn_load_layer_pad},
     {dnn_execute_layer_maximum,     dnn_load_layer_maximum},
     {dnn_execute_layer_math_binary, dnn_load_layer_math_binary},
+    {dnn_execute_layer_math_unary,  dnn_load_layer_math_unary},
 };

tools/python/convert_from_tensorflow.py

@@ -70,8 +70,9 @@ class TFConverter:
         self.converted_nodes = set()
         self.conv2d_scope_names = set()
         self.conv2d_scopename_inputname_dict = {}
-        self.op2code = {'Conv2D':1, 'DepthToSpace':2, 'MirrorPad':3, 'Maximum':4, 'MathBinary':5}
+        self.op2code = {'Conv2D':1, 'DepthToSpace':2, 'MirrorPad':3, 'Maximum':4, 'MathBinary':5, 'MathUnary':6}
         self.mathbin2code = {'Sub':0, 'Add':1, 'Mul':2, 'RealDiv':3, 'Minimum':4}
+        self.mathun2code  = {'Abs':0}
         self.mirrorpad_mode = {'CONSTANT':0, 'REFLECT':1, 'SYMMETRIC':2}
         self.name_operand_dict = {}
 
@@ -286,6 +287,17 @@ class TFConverter:
         np.array([output_operand_index], dtype=np.uint32).tofile(f)
 
 
+    def dump_mathunary_to_file(self, node, f):
+        self.layer_number = self.layer_number + 1
+        self.converted_nodes.add(node.name)
+        i0_node = self.name_node_dict[node.input[0]]
+        np.array([self.op2code['MathUnary'], self.mathun2code[node.op]], dtype=np.uint32).tofile(f)
+        input_operand_index = self.add_operand(i0_node.name, Operand.IOTYPE_INPUT)
+        np.array([input_operand_index], dtype=np.uint32).tofile(f)
+        output_operand_index = self.add_operand(node.name, Operand.IOTYPE_OUTPUT)
+        np.array([output_operand_index],dtype=np.uint32).tofile(f)
+
+
     def dump_layers_to_file(self, f):
         for node in self.nodes:
             if node.name in self.converted_nodes:
@@ -307,6 +319,8 @@ class TFConverter:
                 self.dump_maximum_to_file(node, f)
             elif node.op in self.mathbin2code:
                 self.dump_mathbinary_to_file(node, f)
+            elif node.op in self.mathun2code:
+                self.dump_mathunary_to_file(node, f)
 
 
     def dump_operands_to_file(self, f):

tools/python/convert_header.py

@@ -23,4 +23,4 @@ str = 'FFMPEGDNNNATIVE'
 major = 1
 
 # increase minor when we don't have to re-convert the model file
-minor = 5
+minor = 6