![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F16342273%2F54806d6ca5c3298b740c930a9360103f%2FScreenshot%202025-05-30%20at%2013.49.19.png?generation=1748609376448475&alt=media)

> **Important Notice:**

> **Important Notice:**
> **New Final Submission Deadline ▶️ 10 August 2025 (AOE)**
> We are delighted to let you know that the challenge deadline has been extended! Now you have more time to cook your good stuff! 

> Because the initial Kaggle competition could no longer be re‑opened after its closing date, **a new competition page has been created to host the extended track**:
https://www.kaggle.com/competitions/bin-ego-360-challenge-classification-ext

> All rules, evaluation metrics, and prizes remain unchanged (with an additional prize of registration fee!); only the schedule has shifted, and you can submit via kaggle's sandbox. If you submitted to the first competition, simply re‑upload your final model predictions here before the new deadline. ✨


## Goal of the [BinEgo-360](https://x360dataset.github.io/BinEgo-360/#challenge) Challenge

What does it take for a model to understand a scene—is it enough to see it? To hear it? Or is there more?

We present an exciting new challenge at the frontier of panoptic video understanding. The goal is to explore **multi-perspective learning**—understanding how the fusion of diverse viewpoints and modalities, such as egocentric vision, 360° third-person context, audio, and location cues, can lead to a more complete and holistic representation of complex scenes.

Join us in pushing the boundaries of multi-view learning. Your solution could help shape the future of perception in robotics, AR/VR, and human-centric video intelligence.


~~~~

>**This challenge is only for the ICCV 2025 BinEgo360 workshop and Challenge, and all the prizes will be issued at the workshop**

## Context

This challenge is part of the [BinEgo-360](https://x360dataset.github.io/BinEgo-360/#challenge) workshop. **Winners of the challenge will be invited to jointly write a paper about the workshop and challenge to be included in the ICCV proceedings.** The objective of this track is to predict the scene label (**classification**) for a given video clip, following the scene categories provided in the dataset. For more details about the dataset, tracks, timeline, and submission rules, please visit: https://x360dataset.github.io/BinEgo-360/#challenge.

We primarily incorporate two types of videos:

- **First-person (egocentric) video** captures the visual experience from the capturer's point of view. It includes direct observations of objects, hands, and interactions at close range, often reflecting the actor’s focus and activity. The content is typically constrained by a limited field of view and may include frequent occlusions due to body movement or object handling.
- **360° third-person video** provides an omnidirectional view of the environment from an external perspective. It records the full spatial layout and captures multiple agents within the scene. This perspective remains external to the actor, without access to internal states or viewpoint-specific information.