AI Edge Contest

For the advancement of the cognitive technology of Automobile Driving Recorders

Focusing on "automatic driving,” one of the themes of social implementation in developing innovative AI edge computing technology and development of next generation computing technology, with the aim of discovering excellent technology, advanced ideas and talents, we will hold a contest competing for the accuracy of image recognition technology indispensable for automatic operation.

Application Period
Nov.19, 2018 to Jan.27, 2019

CONTEST THEME

Improvement of recognition technology in car traveling video

OBJECT DETECTION DIVISION SEGMENTATION DIVISION
Assignment Create an algorithm to detect a rectangular region including an object from a vehicle front camera image. Create an algorithm to segment at the pixel level the region corresponding to an object from a vehicle front camera image.
Data
  • (Train/Test) Vehicle front camera image
  • (Train) Labeled rectangular region of an object
  • (Train) Image meta information (route / time of day)
  • (Train/Test) Vehicle front camera image
  • (Train) Split-labeled region corresponding to an object at the pixel level
  • (Train) Image meta information (route / time of day)
Objects Identified Car, Pedestrian, Truck, Bicycle, Signal, Signs Car, Pedestrian, Signal, Lane
Evaluation Method ・Quantitative evaluation based on mAP@IoU=0.75
・Qualitative evaluation based on model document
・Quantitative evaluation based on IoU
・Qualitative evaluation based on model document

OBJECT DETECTION DIVISION

Assignment
Create an algorithm to detect a rectangular region including an object from a vehicle front camera image.
Data
(Train/Test) Vehicle front camera image
(Train) Labeled rectangular region of an object
(Train) Image meta information (route / time of day)
Objects Identified
Car, Pedestrian, Truck, Bicycle, Signal, Signs
Evaluation Method
・Quantitative evaluation based on mAP@IoU=0.75
・Qualitative evaluation based on model documen

SEGMENTATION DIVISION

Assignment
Create an algorithm to segment at the pixel level the region corresponding to an object from a vehicle front camera image.
Data
(Train/Test) Vehicle front camera image
(Train) Split-labeled region corresponding to an object at the pixel level
(Train) Image meta information (route / time of day)
Objects Identified
Car, Pedestrian, Signal, Lane
Evaluation Method
・Quantitative evaluation based on IoU
・Qualitative evaluation based on model document
  • Source data
  • Objects detected bounding box
  • SegmentationCarPedestrianLane

DATA OUTLINE

OBJECT DETECTION DIVISION SEGMENTATION DIVISION
Route
  • Tokyo Route 1 (Shibuya-Haneda))
  • Tokyo Route 2 (Kagurazaka)
  • Saitama Route (Bijogi)
  • Tokyo Route 1 (Shibuya-Haneda)
  • Tokyo Route 2 (Kagurazaka)
  • Saitama Route (Bijogi)
Time of day Morning/ day/night: 3 times (part of the route had daytime only) Morning/ day/night: 3 times (part of the route had daytime only)
Number of Images Approximately 30,000 Approximately 3,000
Photograph Intervals 1 second Tokyo Route: 10 seconds
Saitama Route: 30 seconds
Image Resolution 1936 × 1216 1936 × 1216

OBJECT DETECTION DIVISION

Route
  • Tokyo Route 1 (Shibuya-Haneda)
  • Tokyo Route 2 (Kagurazaka)
  • Saitama Route (Bijogi)
Time of day
Morning/ day/night: 3 times (part of the route had daytime only)
Number of images
Approximately 30,000
Photograph Intervals
1 second
Image Resolution
1936 × 1216

SEGMENTATION DIVISION

Route
  • Tokyo Route 1 (Shibuya-Haneda)
  • Tokyo Route 2 (Kagurazaka)
  • Saitama Route (Bijogi)
Time of day
Morning/ day/night: 3 times (part of the route had daytime only)
Number of images
Approximately 3,000
Photograph Intervals
Tokyo Route: 10 seconds, Saitama Route: 30 seconds
Image Resolution
1936 × 1216

Flow of the Contest

STEP1
Participate in Contest
Confirm the assignment and download data for your local environment analysis.
STEP2
Start analyzing
Access the data offered, confirm the contest rules and create a model.
STEP3
Submit prediction results
Input test data into the model you created to generate a prediction and submit it. Confirm your rank on the scoreboard and challenge as many times as you want while aiming for the top!
STEP4
Winner procedures
The final winners will be awarded at the Awards Ceremony after the contest.

Awards

  • In the awards ceremony scheduled to be held in early March 2019, the winners of each division are awarded the Ministry of Economy, Trade and Industry Commerce Information Policy Bureau award, the top rankers of each division are awarded the Extreme Edge Award, prizes(1st: 500,000 yen(equivalent to $4,400), 2nd: 300,000 yen(equivalent to $2,600), 3rd: 100,000 yen(equivalent to $860), idea award 100,000 yen(equivalent to $860) etc).
  • Qualification to participate in the Japan Automotive AI Challenge that will do test driving with the developed algorithm installed in the cart.

Application Guidelines

Age limit
None
Nationality
Any
Participation fee
Free of charge
Participation style
Individuals or teams

To participate, you must register with the contest implementation platform, “SIGNATE”.

Contest advisories

  • Takeo KanadeUA and Helen Whitaker Professor at Carnegie Mellon University
  • Hiroaki KitanoCEO of Sony Computer Science Laboratories
  • James KuffnerCEO of Toyota Research Institute Advanced Development, Inc.
  • Junichi TsujiiDirector of Artificial Intelligence Research Center at the National Institute of Advanced Industrial Science and Technology
  • Junichi MiyakawaVice President and CTO of SoftBank Corp