]> git.sesse.net Git - ffmpeg/commit
dnn: add tf.nn.conv2d support for native model
authorGuo, Yejun <yejun.guo@intel.com>
Mon, 21 Oct 2019 12:38:03 +0000 (20:38 +0800)
committerPedro Arthur <bygrandao@gmail.com>
Wed, 30 Oct 2019 13:31:55 +0000 (10:31 -0300)
commitdff39ea9f0154ec52b7548b122a4a5332df3c2c6
treea435c8d24d80bc9c38ecd4b93c51113c47acad05
parenta269fa044b1364af1654456c33b7d45407822876
dnn: add tf.nn.conv2d support for native model

Unlike other tf.*.conv2d layers, tf.nn.conv2d does not create many
nodes (within a scope) in the graph, it just acts like other layers.
tf.nn.conv2d only creates one node in the graph, and no internal
nodes such as 'kernel' are created.

The format of native model file is also changed, a flag named
has_bias is added, so change the version number.

Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
libavfilter/dnn/dnn_backend_native.c
libavfilter/dnn/dnn_backend_native_layer_conv2d.c
libavfilter/dnn/dnn_backend_native_layer_conv2d.h
tests/dnn/dnn-layer-conv2d-test.c
tools/python/convert_from_tensorflow.py
tools/python/convert_header.py