deploy_model_on_rasp.py 7.88 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License.  You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied.  See the License for the
# specific language governing permissions and limitations
# under the License.
17
"""
18 19
.. _tutorial-deploy-model-on-rasp:

20 21 22 23 24
Deploy the Pretrained Model on Raspberry Pi
===========================================
**Author**: `Ziheng Jiang <https://ziheng.org/>`_

This is an example of using NNVM to compile a ResNet model and deploy
25
it on Raspberry Pi.
26
"""
27

28 29 30
import tvm
import nnvm.compiler
import nnvm.testing
31 32
from tvm import rpc
from tvm.contrib import util, graph_runtime as runtime
33
from tvm.contrib.download import download_testdata
34 35

######################################################################
36 37
# .. _build-tvm-runtime-on-device:
#
38 39 40
# Build TVM Runtime on Device
# ---------------------------
#
41
# The first step is to build tvm runtime on the remote device.
42 43 44
#
# .. note::
#
45 46 47
#   All instructions in both this section and next section should be
#   executed on the target device, e.g. Raspberry Pi. And we assume it
#   has Linux running.
48
#
49 50 51
# Since we do compilation on local machine, the remote device is only used
# for running the generated code. We only need to build tvm runtime on
# the remote device.
52
#
53
# .. code-block:: bash
54
#
55
#   git clone --recursive https://github.com/apache/incubator-tvm tvm
56 57
#   cd tvm
#   make runtime -j4
58
#
59 60
# After building runtime successfully, we need to set environment varibles
# in :code:`~/.bashrc` file. We can edit :code:`~/.bashrc`
61
# using :code:`vi ~/.bashrc` and add the line below (Assuming your TVM
62
# directory is in :code:`~/tvm`):
63
#
64
# .. code-block:: bash
65
#
66
#   export PYTHONPATH=$PYTHONPATH:~/tvm/python
67
#
68
# To update the environment variables, execute :code:`source ~/.bashrc`.
69 70 71 72

######################################################################
# Set Up RPC Server on Device
# ---------------------------
73 74
# To start an RPC server, run the following command on your remote device
# (Which is Raspberry Pi in our example).
75 76 77 78 79
#
#   .. code-block:: bash
#
#     python -m tvm.exec.rpc_server --host 0.0.0.0 --port=9090
#
80 81
# If you see the line below, it means the RPC server started
# successfully on your device.
82 83 84 85
#
#    .. code-block:: bash
#
#      INFO:root:RPCServer: bind to 0.0.0.0:9090
86
#
87 88

######################################################################
89 90 91
# Prepare the Pre-trained Model
# -----------------------------
# Back to the host machine, which should have a full TVM installed (with LLVM).
92
#
93 94 95
# We will use pre-trained model from
# `MXNet Gluon model zoo <https://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html>`_.
# You can found more details about this part at tutorial :ref:`tutorial-from-mxnet`.
96 97 98 99 100

from mxnet.gluon.model_zoo.vision import get_model
from PIL import Image
import numpy as np

101
# one line to get the model
102 103 104
block = get_model('resnet18_v1', pretrained=True)

######################################################################
105
# In order to test our model, here we download an image of cat and
106
# transform its format.
107
img_name = 'cat.png'
108 109 110
img_path = download_testdata('https://github.com/dmlc/mxnet.js/blob/master/data/cat.png?raw=true',
                             img_name, module='data')
image = Image.open(img_path).resize((224, 224))
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127

def transform_image(image):
    image = np.array(image) - np.array([123., 117., 104.])
    image /= np.array([58.395, 57.12, 57.375])
    image = image.transpose((2, 0, 1))
    image = image[np.newaxis, :]
    return image

x = transform_image(image)

######################################################################
# synset is used to transform the label from number of ImageNet class to
# the word human can understand.
synset_url = ''.join(['https://gist.githubusercontent.com/zhreshold/',
                      '4d0b62f3d01426887599d4f7ede23ee5/raw/',
                      '596b27d23537e5a1b5751d2b0481ef172f58b539/',
                      'imagenet1000_clsid_to_human.txt'])
128 129 130
synset_name = 'imagenet1000_clsid_to_human.txt'
synset_path = download_testdata(synset_url, synset_name, module='data')
with open(synset_path) as f:
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
    synset = eval(f.read())

######################################################################
# Now we would like to port the Gluon model to a portable computational graph.
# It's as easy as several lines.

# We support MXNet static graph(symbol) and HybridBlock in mxnet.gluon
net, params = nnvm.frontend.from_mxnet(block)
# we want a probability so add a softmax operator
net = nnvm.sym.softmax(net)

######################################################################
# Here are some basic data workload configurations.
batch_size = 1
num_classes = 1000
image_shape = (3, 224, 224)
data_shape = (batch_size,) + image_shape

######################################################################
# Compile The Graph
# -----------------
# To compile the graph, we call the :any:`nnvm.compiler.build` function
# with the graph configuration and parameters. However, You cannot to
# deploy a x86 program on a device with ARM instruction set. It means
# NNVM also needs to know the compilation option of target device,
# apart from arguments :code:`net` and :code:`params` to specify the
# deep learning workload. Actually, the option matters, different option
# will lead to very different performance.

######################################################################
161 162 163 164
# If we run the example on our x86 server for demonstration, we can simply
# set it as :code:`llvm`. If running it on the Raspberry Pi, we need to
# specify its instruction set. Set :code:`local_demo` to False if you want
# to run this tutorial with a real device.
165

166 167 168
local_demo = True

if local_demo:
169
    target = tvm.target.create('llvm')
170 171 172
else:
    target = tvm.target.arm_cpu('rasp3b')
    # The above line is a simple form of
173
    # target = tvm.target.create('llvm -device=arm_cpu -model=bcm2837 -target=armv7l-linux-gnueabihf -mattr=+neon')
174

175
with nnvm.compiler.build_config(opt_level=3):
176 177
    graph, lib, params = nnvm.compiler.build(
        net, target, shape={"data": data_shape}, params=params)
178 179 180 181 182 183 184

# After `nnvm.compiler.build`, you will get three return values: graph,
# library and the new parameter, since we do some optimization that will
# change the parameters but keep the result of model as the same.

# Save the library at local temporary directory.
tmp = util.tempdir()
185 186
lib_fname = tmp.relpath('net.tar')
lib.export_library(lib_fname)
187 188 189 190 191 192 193

######################################################################
# Deploy the Model Remotely by RPC
# --------------------------------
# With RPC, you can deploy the model remotely from your host machine
# to the remote device.

194 195 196 197 198 199 200 201
# obtain an RPC session from remote device.
if local_demo:
    remote = rpc.LocalSession()
else:
    # The following is my environment, change this to the IP address of your target device
    host = '10.77.1.162'
    port = 9090
    remote = rpc.connect(host, port)
202 203 204

# upload the library to remote device and load it
remote.upload(lib_fname)
205
rlib = remote.load_module('net.tar')
206 207

# create the remote runtime module
208
ctx = remote.cpu(0)
209
module = runtime.create(graph, rlib, ctx)
210 211
# set parameter (upload params to the remote device. This may take a while)
module.set_input(**params)
212 213 214 215 216
# set input data
module.set_input('data', tvm.nd.array(x.astype('float32')))
# run
module.run()
# get output
217
out = module.get_output(0)
218 219 220
# get top1 result
top1 = np.argmax(out.asnumpy())
print('TVM prediction top-1: {}'.format(synset[top1]))