[FFmpeg-cvslog] dnn_backend_native_layer_mathunary: add atan support

Ting Fu git at videolan.org
Thu Jun 25 03:47:25 EEST 2020


ffmpeg | branch: master | Ting Fu <ting.fu at intel.com> | Thu Jun 18 17:15:35 2020 +0800| [13f5613e684f65e625cb287f85c3410f63efbf1c] | committer: Guo, Yejun

dnn_backend_native_layer_mathunary: add atan support

It can be tested with the model generated with below python script:

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.atan(x)
x2 = tf.divide(x1, 3.1416/4) # pi/4
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 at intel.com>
Signed-off-by: Guo Yejun <yejun.guo at intel.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=13f5613e684f65e625cb287f85c3410f63efbf1c
---

 libavfilter/dnn/dnn_backend_native_layer_mathunary.c | 4 ++++
 libavfilter/dnn/dnn_backend_native_layer_mathunary.h | 1 +
 tools/python/convert_from_tensorflow.py              | 2 +-
 tools/python/convert_header.py                       | 2 +-
 4 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/libavfilter/dnn/dnn_backend_native_layer_mathunary.c b/libavfilter/dnn/dnn_backend_native_layer_mathunary.c
index d130058546..42615c43d5 100644
--- a/libavfilter/dnn/dnn_backend_native_layer_mathunary.c
+++ b/libavfilter/dnn/dnn_backend_native_layer_mathunary.c
@@ -100,6 +100,10 @@ int dnn_execute_layer_math_unary(DnnOperand *operands, const int32_t *input_oper
         for (int i = 0; i < dims_count; ++i)
             dst[i] = acos(src[i]);
         return 0;
+    case DMUO_ATAN:
+        for (int i = 0; i < dims_count; ++i)
+            dst[i] = atan(src[i]);
+        return 0;
     default:
         return -1;
     }
diff --git a/libavfilter/dnn/dnn_backend_native_layer_mathunary.h b/libavfilter/dnn/dnn_backend_native_layer_mathunary.h
index f146248567..13fa33178a 100644
--- a/libavfilter/dnn/dnn_backend_native_layer_mathunary.h
+++ b/libavfilter/dnn/dnn_backend_native_layer_mathunary.h
@@ -36,6 +36,7 @@ typedef enum {
     DMUO_TAN = 3,
     DMUO_ASIN = 4,
     DMUO_ACOS = 5,
+    DMUO_ATAN = 6,
     DMUO_COUNT
 } DNNMathUnaryOperation;
 
diff --git a/tools/python/convert_from_tensorflow.py b/tools/python/convert_from_tensorflow.py
index 78297e48a9..b90c31c495 100644
--- a/tools/python/convert_from_tensorflow.py
+++ b/tools/python/convert_from_tensorflow.py
@@ -72,7 +72,7 @@ class TFConverter:
         self.conv2d_scopename_inputname_dict = {}
         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, 'Sin':1, 'Cos':2, 'Tan':3, 'Asin':4, 'Acos':5}
+        self.mathun2code  = {'Abs':0, 'Sin':1, 'Cos':2, 'Tan':3, 'Asin':4, 'Acos':5, 'Atan':6}
         self.mirrorpad_mode = {'CONSTANT':0, 'REFLECT':1, 'SYMMETRIC':2}
         self.name_operand_dict = {}
 
diff --git a/tools/python/convert_header.py b/tools/python/convert_header.py
index 4a8e44b4aa..73cf23bf53 100644
--- a/tools/python/convert_header.py
+++ b/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 = 11
+minor = 12



More information about the ffmpeg-cvslog mailing list