NCVPRIPG 2023 Challenge on Categorizing Missing

Traffic Signs from Contextual Cues (C4MTS)

Get Started

Important Updates

  • Congratulations Sandeep Nagar for Winner award, and Suman Paul Choudhury , Rohit Singh for the Runner-up award.
  • NCVPRIPG'23 early bird registration deadline is extended to 2 July'23.
  • C4MTS submission deadline is extended to 1 July'23 11:55 p.m. (no more extensions after this)
  • Submission deadline is extended to 25 June 2023.
  • Training dataset and test images are now available at .
  • Refer to the README in the test data folder for submission guidelines.
  • The Submission deadline is extended to 15 June 2023 at 11:55 p.m. sharp.
  • Task 1 baseline code will soon be available at
  • Task 2 baseline code is now available at:
  • Acknowledgement: Thanks to Anandita Jamwal (PhD Student, IIT Mandi) and Vibhu Gupta (BTech Student, IIT Mandi) for setting up the baselines.
  • The mAP scores and top-1 accuracy of the baselines will soon be shared with the leaderboard.


The international Autonomous Vehicle (AV) market is estimated to reach around two thousand billion by 2030. Millions of lives are lost yearly in road accidents, and traffic violations cause a significant percentage of AV accidents. The traffic signs are generally installed at the side of the road to control traffic flow or pass information about the road environment to Vulnerable Road Users (VRUs). Often, the information is also available in the form of cues present in the context around the traffic signs in the cues away from it, which we refer to as contextual cues.


The C4MTS challenge of Categorizing Missing Traffic Signs provides 200 real scenes from Missing Traffic Sign Video Dataset (MTSVD) uniformly distributed over four types of missing traffic signs: left-hand-curve, right-hand-curve, gap-in-median, and side-road-left. The traffic signs are common but individually observed with contextual cues. In the given examples, contextual cues like rumble strips, side roads, etc., are present, but the traffic signs are missing due to improper planning, lack of budget, etc. To learn the relationship between the traffic signs and the contextual cues, 2000 training images, each containing one of the four traffic signs (commonly and individually visible with contextual cues) and corresponding bounding boxes, are also provided. Two tasks are proposed for the challenge: i) Object Detection, wherein model training happens using bounding box annotations provided with the data, and ii) Missing Traffic Sign Scene Categorization, wherein model training happens using road scene images with in-painted traffic signs provided with the data. The models for the second task will be tested on a mixture of images with in-painted traffic signs and images with missing traffic signs to encourage inpainting agnostic real-life solutions.

Important Dates

Topic Dates
Release of Training Dataset May 17, 2023
Opening Date for Submission to Challenges June 09, 2023
Release of Test Set Images June 09, 2023
Closing Date for Submission to Challenges July 01, 2023
Announcement of Challenge Winners July 02, 2023
Registration Deadline for Challenge Attendees June 25, 2023

Note: Compulsory for winners to attend the NCVPRIPG'23, registeration will be waived off for winners.


Interested in participation? Please Click Here to register.

Participation Rules

  • Participants must use data only for the competition, and must not share the data with anyone else as it is only for research purposes and has license involved while downloading form IDD website
  • Participants are not allowed to use the dataset outside the competition. The dataset for task 1 may be used for task 2 if participants want to.
  • It is compulsory for the winner and the runner-up teams/individuals to register and attend NCVPRIPG'23.
  • Any individual or group can use email-id to create a participation id for the competition.
  • Making multiple ids of the same group or individual is strictly prohibited.
  • No restriction exists on the number of groups from an institute/organization, but common participants in the groups (from the same/different institute/organization) are not allowed.
  • The groups or individuals must upload the obtained results (in the specific format) to the challenge website, which organizers will verify.
  • Scripts will calculate the scores for the proposed task and display the results on the leaderboard.
  • The group members or individuals can update the results on the leaderboard, with a limit of 3 times per day per group (or individual in case of no group), after the test data release and before the end of the competition.
  • To verify the results and ensure fair participation, the participants or top teams will be told to reproduce them at the event (or before) using the trained models. If the submitted results are mismatched, the participants will be disqualified, and their results will be removed from the leaderboard.
  • The decision for awards by organizers will be final

Challenge Leaderboard

For Task 1, the scores are as follows:

S.No Team Name/ Task1Method mAP
01 Baseline 88 %
02 Sahajeevis 84 %
03 IAMGROOT 90.00 %

For Task 2, the scores are as follows:

S.No Team Name/ Task2Method Top-1 Accuracy
01 Baseline 29 %
02 Sahajeevis 51.50 %
03 IAMGROOT 60.50 %


IHub - Data at IIIT Hyderabad is sponsoring the following awards for the winners of the Challenge:

  • Winner: INR 5,000
  • Runner-up: INR 3,000
  • Noteworthy approach : INR 2,000


Rohit Saluja

IIT Mandi

Varun Gupta

IIIT Hyderabad

Anbumani Subramanian


Prof. C V Jawahar

IIIT Hyderabad


For Registration, Please Click Here

For any enquiry, please contact