CHALLENGE

Person attribute recognition (PAR) is crucial for city surveillance as it enables the identification and tracking of individuals across multiple cameras. It also gives the system the ability to retrieve instances that have specific attributes, a crucial requirement in surveillance applications. Typical attributes include gender, clothing style, carrying items, etc, which provide high-level semantic information. Existing standard datasets like PETA, Market 1501, and PA100K lack attribute classes relevant to Indian attire, such as kurta, salwar, dupatta, and saree, etc. that are essential for the Indian context. Several solutions are also available in the literature which gives good results on these datasets. But these solutions are not directly applicable to the Indian scenario where there is a change in skin color, dressing style, etc. Even the class information needs to be modified to suit the Indian scenario. The proposed challenge addresses these gaps by focusing on detecting person attributes explicitly tailored for the Indian scenario, enhancing the accuracy and relevance of attribute recognition in smart city environments. We will be providing a sample dataset with the intent of sensitizing the participants to the Indian scenario. Participants are encouraged to enhance the dataset to meet their training requirements in a suitable manner.

This is the second edition of our Person Attribute Recognition (PAR) challenge, building on the foundation of the previous iteration with a more challenging dataset. This time, we are introducing complex scenarios, including partial occlusion and variations in appearance, to better reflect practical surveillance conditions. Additionally, we are imposing constraints on accuracy, inference time, and model size to encourage the development of efficient and scalable solutions suitable for real-time smart city surveillance.

SIGNIFICANCE

In today’s digital era, the presence of a PAR system holds immense importance in both civilian and military applications. Such systems allow for the identification of individuals based on specific attributes within surveillance videos, a capability crucial for security purposes. With the widespread deployment of surveillance cameras in smart cities, as well as in residential and commercial settings, the need for robust PAR solutions is more pressing than ever. While existing solutions may perform well on curated datasets, the real challenge lies in developing systems that can handle the complexity of real-world data. This challenge serves as an opportunity for researchers to collaborate and create practical, effective PAR solutions tailored to the Indian context, leveraging the collective knowledge and expertise of the research community.

RULES FOR PARTICIPATION

1. Shared annotated dataset is the legal property of Vehant and is shared strictly for usage in challenge alone. Please refrain from using it for any other project/ publication/ application. Also ensure data security and delete the data after the challenge completion.
2. Only one member from the team has to fill the registration form for the challenge. Multiple entries of the same group or individual is strictly prohibited.
3. The approach will be evaluated based on the label-based mean accuracy (mA), time and space complexity on a private test set.
4. 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.
5. Participants may be required to submit the source code of their models, along with documentation describing the architecture, training procedure, and any additional pre-processing or post-processing steps.
6. Attending the conference (NCVPRIPG’25) will be highly encouraged. At least one person from winning teams of the challenge must attend the challenge session in person.

Interested In Participation? Please Click Here To Register

 

IMPORTANT DATES

Event	Date
Registration opening and launch of challenge website	April 15, 2025
Release of Training Dataset	April 30, 2025
Opening Date for Submission to Challenges	May 15, 2025
Release of Test Set Images	May 15, 2025
Closing Date for Submission to Challenges	June 5, 2025
Winner announcement	June 25, 2025

 

AWARDS

Position	Prize
1st Place (Winner)	25,000
2nd Place	20,000

 

CONTACT 

Organizers : Dr. Renu M. Rameshan, Dr. Shikha Gupta, Anurag Bajpai 

For any query please contact: contest@vehant.com