Skip to content
Branch: master
Find file History
Type Name Latest commit message Commit time
Failed to load latest commit information.
tf-cpp Add tf-cpp accuracy for 100% cropping. Mar 27, 2019
tflite Add bonus model (resnet) to tflite. Mar 27, 2019 Use MobileNet when referring to specific MLPerf instance. Mar 27, 2019

compatibility License

MLPerf Inference - Object Classification - MobileNet

NB: MLPerf Inference v0.5 uses MobileNets-v1-1.0-224 (called MobileNet in what follows).

Table of contents

  1. Installation
  2. Benchmarking


Debian (tested with Ubuntu v18.04 and v16.04)

Install common tools and libraries

$ sudo apt install autoconf autogen libtool zlib1g-dev
$ sudo apt install gcc g++ git wget
$ sudo apt install libblas-dev liblapack-dev

Install Python, pip, SciPy and CK

$ sudo apt install python3 python3-pip
$ sudo python3 -m pip install scipy
$ sudo python3 -m pip install ck

NB: CK also supports Python 2.

[Optional] Install Android SDK and NDK

You can optionally target Android API 23 (v6.0 "Marshmallow") devices using the --target_os=android23-arm64 flag (or similar), when using the TensorFlow Lite benchmark (recommended) and TensorFlow (C++) benchmark (not recommended).

On Debian Linux, you can install the Android SDK and the Android NDK as follows:

$ sudo apt install android-sdk
$ adb version
Android Debug Bridge version 1.0.36
Revision 1:7.0.0+r33-2
$ sudo apt install google-android-ndk-installer

NB: On Ubuntu 18.04, NDK r13b gets installed. On Ubuntu 16.04, download NDK r18b and extract it into e.g. /usr/local. NDK r18c only supports LLVM, which currently requires a CK quirk to work properly (removing a dependency on from

Install CK workflows for MLPerf

Pull CK repositories

$ ck pull repo:ck-mlperf

NB: Transitive dependencies include repo:ck-tensorflow.

Install a small dataset (500 images)

$ ck install package:imagenet-2012-val-min 

NB: ImageNet dataset descriptions are in repo:ck-env.

Install the full dataset (50,000 images)

$ ck install package:imagenet-2012-val

NB: If you already have the ImageNet validation dataset downloaded in a directory e.g. $HOME/ilsvrc2012-val/, you can simply detect it as follows:

$ ck detect soft:dataset.imagenet.val --full_path=$HOME/ilsvrc2012-val/ILSVRC2012_val_00000001.JPEG


You can benchmark MobileNet using one of the available options:

You can’t perform that action at this time.