it can be tested with model file generated from above python script: import tensorflow as tf import numpy as np import imageio in_img = imageio.imread('input.jpg') 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') z1 = 0.5 + 0.3 * x z2 = z1 * 4 z3 = z2 - x - 2.0 y = tf.identity(z3, 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))
Unified · Split
| @@ -120,6 +120,19 @@ int dnn_execute_layer_math_binary(DnnOperand *operands, const int32_t *input_ope · 119 unmodified lines | ||
| 120 | 120 | } |
| 121 | 121 | } |
| 122 | 122 | return 0; |
| 123 | + case DMBO_MUL: | |
| 124 | + if (params->input0_broadcast || params->input1_broadcast) { | |
| 125 | + for (int i = 0; i < dims_count; ++i) { | |
| 126 | + dst[i] = params->v * src[i]; | |
| 127 | + } | |
| 128 | + } else { | |
| 129 | + const DnnOperand *input1 = &operands[input_operand_indexes[1]]; | |
| 130 | + const float *src1 = input1->data; | |
| 131 | + for (int i = 0; i < dims_count; ++i) { | |
| 132 | + dst[i] = src[i] * src1[i]; | |
| 133 | + } | |
| 134 | + } | |
| 135 | + return 0; | |
| 123 | 136 | default: |
| 124 | 137 | return -1; |
| 125 | 138 | } |
| @@ -33,6 +33,7 @@ · 32 unmodified lines | ||
| 33 | 33 | typedef enum { |
| 34 | 34 | DMBO_SUB = 0, |
| 35 | 35 | DMBO_ADD = 1, |
| 36 | + DMBO_MUL = 2, | |
| 36 | 37 | DMBO_COUNT |
| 37 | 38 | } DNNMathBinaryOperation; |
| 38 | 39 | |
| @@ -71,7 +71,7 @@ class TFConverter: · 70 unmodified lines | ||
| 71 | 71 | self.conv2d_scope_names = set() |
| 72 | 72 | self.conv2d_scopename_inputname_dict = {} |
| 73 | 73 | self.op2code = {'Conv2D':1, 'DepthToSpace':2, 'MirrorPad':3, 'Maximum':4, 'MathBinary':5} |
| 74 | − self.mathbin2code = {'Sub':0, 'Add':1} | |
| 74 | + self.mathbin2code = {'Sub':0, 'Add':1, 'Mul':2} | |
| 75 | 75 | self.mirrorpad_mode = {'CONSTANT':0, 'REFLECT':1, 'SYMMETRIC':2} |
| 76 | 76 | self.name_operand_dict = {} |
| 77 | 77 | |
| @@ -309,6 +309,8 @@ class TFConverter: · 231 unmodified lines | ||
| 309 | 309 | self.dump_mathbinary_to_file(node, f) |
| 310 | 310 | elif node.op == 'Add': |
| 311 | 311 | self.dump_mathbinary_to_file(node, f) |
| 312 | + elif node.op == 'Mul': | |
| 313 | + self.dump_mathbinary_to_file(node, f) | |
| 312 | 314 | |
| 313 | 315 | |
| 314 | 316 | def dump_operands_to_file(self, f): |
| @@ -23,4 +23,4 @@ str = 'FFMPEGDNNNATIVE' · 22 unmodified lines | ||
| 23 | 23 | major = 1 |
| 24 | 24 | |
| 25 | 25 | # increase minor when we don't have to re-convert the model file |
| 26 | −minor = 2 | |
| 26 | +minor = 3 | |