The Persistent Scene Change Detection
and Localization (PSCDL 2026)

CHALLENGE

The Persistent Scene Change Detection and Localization (PSCDL 2026) challenge addresses the problem of identifying and localizing long-duration scene modifications in fixed-camera urban surveillance videos. While existing video analytics systems primarily focus on short-term events such as motion or intrusion, real-world urban management often requires detecting persistent changes such as unattended objects, debris accumulation, graffiti, illegal dumping, or temporary encroachments that remain over extended periods. These sustained modifications, though not immediately hazardous, can impact public safety, accessibility, and civic order if left unaddressed. Detecting them reliably is challenging due to illumination variations, weather conditions, dynamic backgrounds, and the need for robust temporal reasoning to distinguish transient motion from persistent presence.

In PSCDL 2026, participants are provided with video clips containing an initial clean baseline segment followed by the introduction of a persistent change and are tasked with both detecting the occurrence of the modification and generating a pixel-level binary mask for accurate localization. Submissions will be evaluated using pixel-level segmentation metrics such as Precision, Recall, and F1-score, with the goal of establishing a benchmark that advances practical, deployment-ready solutions for long-term scene monitoring in urban environments.

HIGHLIGHTS

  • Data: Video clips begin with normal scene activity; later, a new object is introduced and remains visible for an extended duration.

  • Goal: Detect persistent scene changes and generate a pixel-level binary mask highlighting the change.

  • Evaluation: Performance is measured using pixel-level Precision, Recall, and F1-score.

  • Team: A team can consist of a minimum of 1 and a maximum of 4 members.

Why Compete?

Top-performing participants may receive an internship opportunity at Vehant Research Lab, working on real-world computer vision systems from traditional ML to Vision-Language Models (VLMs).

 

IMPORTANT DATES

Event Date

Registration opening and launch of challenge website

February 27, 2026

Release of dataset

March 30, 2026

Release of test set 

April 13, 2026

Opening date for submission to challenge

April 20, 2026

Closing date for submission to challenge

June 5, 2026

Winner announcement

June 20, 2026

AWARDS

Position Prize
Winner 25,000
Runner-up 20,000

REGISTRATION

CONTACT 

Organizers:

Dr. Shikha Gupta, Shivam Nigam, Md. Rizwan Ahmad,  Abhijith P Mahadevan

For any query please contact: contest@vehant.com