My environment is based on：
tensorrt 5.1.5
ubuntu 16.0.4
tensorflow 1.13
CUDA10
Cudnn 7.5.0
In this link I see tensorrt supports tf.transpose,https://docs.nvidia.com/deeplearning/sdk/tensorrt-support-matrix/index.html
This is the list of tensorrt support op of tensorflow:(copy from the link)
SoftMax
Note: If the input to a TensorFlow SoftMax op is not NHWC, TensorFlow will automatically insert a transpose layer with a non-constant permutation, causing the UFF converter to fail. It is therefore advisable to manually transpose SoftMax inputs to NHWC using a constant permutation.
Softplus
Softsign
Transpose
But,when i test the tf.transpose,it got a erroe result:
[TensorRT] ERROR: UffParser: Parser error: output: Order size is not matching the number dimensions of TensorRT
it is my code of test:
import tensorflow.contrib.slim as slim
import tensorflow as tf
model_path = "./transpose_cancat.pb"
data_format = "NCHW"
inputs = tf.placeholder(dtype=tf.float32, shape=(1, 3, 416, 736))
conv1a = slim.conv2d(inputs, 16, [3,3],2,data_format=data_format)
conv2 = tf.transpose(conv1a, perm=[0,2,3,1])
conv2 = slim.conv2d(conv2, 32, [3,3],1)
conv2 = tf.transpose(conv2, perm=[0,3,1,2])
result = tf.concat([conv1a, conv2], 1, name='output')
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
print("1, 2 :", conv1a.shape,conv2.shape)
graphdef = tf.get_default_graph().as_graph_def()
frozen_graph = tf.graph_util.convert_variables_to_constants(sess, graphdef, ['output'])
tf_model = tf.graph_util.remove_training_nodes(frozen_graph)
with tf.gfile.FastGFile(model_path, "wb") as f:
f.write(tf_model.SerializeToString())