Skip to content

Latest commit

 

History

History
63 lines (49 loc) · 3.33 KB

image_classification.md

File metadata and controls

63 lines (49 loc) · 3.33 KB

Image Classification GoogLeNet v1 Demo

Introduction

Image classification is one of the most common benchmarks for machine learning. This tutorial shows you how to launch the image classification GoogLeNet v1 8-bit demo from the Test Drive environment. Once the demo is started, you will be able to view the demo and monitor demo performance from any internet connected web browser.

This demo is meant to use ImageNet ILSVRC2012 validation files. Files should be located in: /home/centos/xfdnn_18_03_19/imagenet_val with the following naming convention: ILSVRC2012_val_.JPEG where <IMAGE_ID> starts at 00000000. Instructions for downloading ILSVRC2012 files can be found here: http://www.image-net.org/download-imageurls

If you want to try it without downloading the dataset, you can insert you own files and simply name them:

  • ILSVRC2012_val_00000000.JPEG
  • ILSVRC2012_val_00000001.JPEG
  • ILSVRC2012_val_00000002.JPEG
  • etc.

The demo should start as long as there is at least one image to classify. Note that the demo webpage shows the correct answer above each image and that is indexed by the ID of the ILSVRC2012 images, so if the images provided are not part of the ILSVRC2012 dataset, there may be a mismatch in the reported labels.

For instructions on launching and connecting to aws instances, see here.

  1. Connect to F1

  2. Navigate to /home/centos/xfdnn_18_03_19/caffe/

    $ ls
classification.bin            libs                  run_mp_conv_xdnn.sh           servergui
data                          models                run_mp_fc_xdnn.sh             start_caffe_docker.sh
examples                      run_common.sh         run_mp_fpga_flow              xdnn_scheduler
exec_caffe_docker.sh          run_cpu_env.sh        run_resnet_16b.sh             xlnx-docker
imagenet                      run_demo_gui.sh       run_resnet_8b.sh              xlnx-xdnn-f1
kernelSxdnn_hw_f1_16b.xclbin  run_demo.sh           sdaccel.ini
kernelSxdnn_hw_f1_8b.xclbin   run_googlenet_16b.sh  sdaccel_profile_summary.csv
kill_demo.sh                  run_googlenet_8b.sh   sdaccel_profile_summary.html

  3. Execute ./start_docker.sh to enter application docker and navigate to /opt/caffe/demo/

    $ ./start_docker.sh
# cd caffe/demo/
# ls
kill_demo.sh  run_demo_gui.sh      run_mp_fc_xdnn.sh  sdaccel.ini                  sdaccel_profile_summary.html
run_demo.sh   run_mp_conv_xdnn.sh  run_mp_fpga_flow   sdaccel_profile_summary.csv  servergui

  4. Execute the ./run_demo.sh script to start the demo

    # ./run_demo.sh
Starting demo...
Starting producer...
Starting web interface...

    Start up of the demo will take a few minutes, but once it's complete, the console will start displaying numbers.

  5. From your host machine: the demo will display at the following web address: http://<yourpublicdns>.compute-1.amazonaws.com:8998/static/www/xdnn1.html

    To get your <yourpublicdns>.compute-1.amazonaws.com address refer to the launching and connecting instructions.

    From your browser you will see the running Image Classification Demo:

  6. To stop the demo, type Ctrl+C and execute ./kill_demo.sh