README.md

---
dataset_info:
  features:
  - name: idx
    dtype: int64
  - name: video_path
    dtype: string
  - name: question
    dtype: string
  - name: choices
    struct:
    - name: a
      dtype: string
    - name: b
      dtype: string
    - name: c
      dtype: string
    - name: d
      dtype: string
  - name: answer
    sequence: string
  - name: choice_type
    dtype: string
  - name: video_source
    dtype: string
  - name: video_type
    dtype: string
  - name: frame_number
    dtype: int64
  - name: video_time
    dtype: float64
  - name: fps
    dtype: float64
  - name: box
    sequence:
      sequence: int64
  - name: mask
    list:
    - name: counts
      dtype: string
    - name: size
      sequence: int64
  - name: point
    sequence:
      sequence: int64
  splits:
  - name: test
    num_bytes: 4578328
    num_examples: 3277
  download_size: 2933575
  dataset_size: 4578328
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
task_categories:
- video-text-to-text
---

# EOC-Bench : Can MLLMs Identify, Recall, and Forecast Objects in an Egocentric World?

<div align=left>

[![arXiv preprint](https://img.shields.io/badge/arxiv-2506.05287-ECA8A7?logo=arxiv)](https://arxiv.org/abs/2506.05287) 
[![GitHub](https://img.shields.io/badge/%20Git%20Hub-Code-yellow)](https://github.com/alibaba-damo-academy/EOCBench/)
[![Project Page](https://img.shields.io/badge/🌐%20Project-Page-9DC3E6)](https://circleradon.github.io/EOCBench/)
[![Learderboard](https://img.shields.io/badge/🏆%20Leaderboard-Page-96D03A)](https://circleradon.github.io/EOCBench/#leaderboard)


## 🔍 Overview
we introduce <strong>EOC-Bench</strong>, an innovative benchmark designed to systematically evaluate object-centric embodied cognition in dynamic egocentric scenarios.
Specially, <strong>EOC-Bench</strong> features 3,277 meticulously annotated QA pairs categorized into three temporal categories: Past, Present, and Future, covering 11 fine-grained evaluation  dimensions and 3 visual object referencing types.
To ensure thorough assessment, we develop a  mixed-format human-in-the-loop annotation framework with four types of questions and design a novel multi-scale temporal accuracy metric for open-ended temporal evaluation. 

<p align="center">
    <img src="https://cdn-uploads.huggingface.co/production/uploads/64a3fe3dde901eb01df12398/gJb0lE0mi6EskZQ8H0Qsm.png" width="100%" style="margin-bottom: 0.2;"/>
<p>

## 📚 Tasks Definition
EOC-Bench structures questions into three temporally grounded categories: **Past, Present, and Future**, with a total of **11** categories.

![data.png](https://cdn-uploads.huggingface.co/production/uploads/64a3fe3dde901eb01df12398/wDwXgMA6UNyvtdWhqqzq9.png)

### 📈 Evaluation
Please see our [GitHub](https://github.com/alibaba-damo-academy/EOCBench/).