Low-Power Image Recognition Challenge (LPIRC 2019)

 

CVPR Workshop Low-Power Image Recognition Challenge (LPIRC)

Held as part of Computer Vision and Pattern Recognition Conference (CVPR 2019)

Long Beach, CA, USA

17 June 2019 | Long Beach, California, USA

Recent publication:
Low-Power Computer Vision: Status, Challenges, and Opportunities
IEEE Journal on Emerging and Selected Topics in Circuits and Systems (Volume: 9, Issue: 2, June 2019)

Read about previous LPIRC competitions

 

Winning Teams from LPIRC 2019

The 2019 IEEE Low-Power Image Recognition Challenge (LPIRC) concluded successfully on 17 June. This year a record number of 22 teams participated and they submitted 234 solutions. Most solutions are significantly better than the solutions in 2018. This challenge has two different tracks: object detection and image classification.

In the object detection track, the first prize is obtained by the Amazon team; the members are Tao Sheng, Peng Lei, Zhuo Deng, Lu Xia, Yuyin Sun, Feng Ye, Lijun Li, Yang Liu, Cheng Hao Kuo, Ning Zhou, Jim Thomas, Kah Kuen Fu, and Dan Chernikoff. The second prize goes to the Tsinghua University China team composed of Wenshuo Li, Chengyu Shi, Zhihang Li, and Teng Xi. The third prize is received by the Qualcomm team, including Jiancheng Lyu, Shuai Zhang, Mengran Gou, Yingyong Qi, and Ning Bi.

In the image classification track, the first Prize goes to Deepinali from Alibaba. The team members are Chen Hesen, Wang Zhibin, Dou Zesheng, Sun Xiuyu, and Li Hao. Expasoft received the second prize and the team includes Sergey Alyamkin, Alexandr Goncharenko, David Svitov, Andrey Denisov, and Timofey Naumenko. The third prize is the MIT HAN Lab from Massachusetts Institute of Technology. The members are Han Cai, Tianzhe Wang, Zhanghao Wu, Kuan Wang, Ji Lin, Song Han.

LPIRC started in 2015 with the aim to identify the technologies for computer vision using energy efficiently. IEEE Rebooting Computing has been the primary sponsor and provides administrative support. In 2019, sponsors are Google, Mediatek, Xilinx, and IEEE Circuits and Systems Society. Past sponsors include Facebook, Nvidia, IEEE GreenICT, IEEE Council of Design Automation, IEEE Council on Superconductivity. The 2019 LPIRC is organized by Purdue University, Duke University, University of North Carolina Chapel Hill. Google has been providing a unified latency metric and an evaluation platform for the online track (also called On-Device Visual Intelligence Challenge) since 2018.

Thanks to all participants, organizing committee members, and student assistants!

 

LPIRC 2019 Gallery

LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo LPIRC photo

 

LPIRC 2019 Overview

This workshop will extend the successes of LPIRC in the past four years identifying the best vision solutions that can simultaneously achieve high accuracy in computer vision and energy efficiency. Since the first competition held in 2015, the winners’ solutions have improved 24 times in the ratio of accuracy divided by energy.

LPIRC 2019 will continue the online + onsite options for participants. One online track with pre-selected hardware allows participants to submit their solutions multiple times without the need of traveling. One onsite track allows participants to bring their systems to CVPR.

  • Online Track (objection detection): Tensorflow Model. Submission dates: 2019/05/15-2019/06/10
  • Onsite Track (objection detection): No hardware or software restriction. Bring your system to CVPR.

More details of the tracks are described below.

For the onsite track, Xilinx Ultra96 + PYNQ framework http://www.pynq.io/community.html is an option available to participants. Xilinx will provide support and possible HW donation. People interested in this opportunity may contact Naveen Purushotham (npurusho@xilinx.com) for more details.

All participants must register before 2019/05/30. Registration open 2019/03/15 at https://lpirc.ecn.purdue.edu/.

To win a prize in 2019 LPIRC, a solution must be better than the state of the art. Please note that this is a new requirement in 2019. The requirement in each track is shown below. If a solution’s score is below the threshold performance, it cannot win. If a solution’s score is better, it still has to beat the other solutions submitted to 2019 LPIRC.

Online Track: mAP 0.14684
Onsite Track: mAP / energy (WH) 0.4446

LPIRC 2019 is a one-day workshop and will feature three speakers, a session on future low-power computer vision, and the onsite competition. The program for LPIRC is shown below.

2019 winners will be invited to speak in a workshop co-located with 2019 ICCV.

09:30-09:40 Welcome by Organizers and Summary of Online Challenge
09:40-10:30 Speech by the Competition Winners (2018 LPIRC-II)
10:30-11:10 Invited Speaker: Rethinking the Computations in Computer Vision (and the Hardware that Computes Them), Kurt Keutzer, Berkeley
11:10-12:00 Invited Speakers
12:00-12:40 Lunch Break
12:40-14:40 Invited Speakers
14:40-17:00 Onsite Competition
17:00-17:30 Discussion: Future Low-Power Computer Vision

If you are interested in discussion of future competitions of low-power computer vision, please fill out this form.

For inquiries, please contact rcinfo@ieee.org.

 

LPIRC 2019 Links

LPIRC Onsite Track

LPIRC Online Track

Abstracts of Invited Talks

 

Sponsors

IEEE Rebooting ComputingGoogleIEEE Circuits and Systems SocietyXilinxMediaTek

 

LPIRC 2019 is sponsored by Xilinx, Google, Mediatek, IEEE Circuits and Systems Society.

LPIRC 2015-2018 sponsors include Google, Facebook, Nvidia, Xilinx, Mediatek, IEEE Circuits and Systems Society, IEEE Council on Electronic Design Automation, IEEE GeenICT, and IEEE Council on Superconductivity. IEEE Rebooting Computing is the founding sponsor of LPIRC.

Sponsorship Opportunities Available

 

Organizing Committee

  • Yung-Hsiang Lu (Purdue)
  • Yiran Chen (Duke)
  • Alex Berg (UNC)
  • Bo Chen (Google)

 

Previous Competitions