Datasets:
license: mit
task_categories:
- video-classification
tags:
- temporal-reasoning
- video-understanding
- benchmark
- vision-language
dataset_info:
features:
- name: relative_path
dtype: string
- name: file
struct:
- name: bytes
dtype: binary
- name: path
dtype: string
splits:
- name: train
num_bytes: 3012200454
num_examples: 595
download_size: 3010737092
dataset_size: 3012200454
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
SpookyBench: A Benchmark for Purely Temporal Video Understanding
SpookyBench is a novel benchmark dataset designed to evaluate the ability of video-language models (VLMs) to understand purely temporal patterns, independent of spatial cues. The dataset consists of 451 videos across four categories: Text, Object Images, Dynamic Scenes, and Shapes. Each video appears as random noise in individual frames, but reveals meaningful content (words, objects, etc.) when viewed as a temporal sequence. This design exposes a critical limitation in current VLMs, which often heavily rely on spatial information and struggle to extract meaning from purely temporal sequences.
Paper: Time Blindness: Why Video-Language Models Can't See What Humans Can?
Project Website: https://timeblindness.github.io/
The dataset contains 451 videos distributed as follows:
| Category | Total Videos | Description |
|---|---|---|
| Text | 210 (46.6%) | English words encoded through temporal noise patterns |
| Object Images | 156 (34.6%) | Single objects encoded using temporal animation |
| Dynamic Scenes | 57 (12.6%) | Video depth maps with temporal motion patterns |
| Shapes | 28 (6.2%) | Geometric patterns encoded through temporal sequences |
| Total | 451 | Comprehensive temporal understanding evaluation |
Download: You can download the dataset from Hugging Face using the following command:
wget https://huggingface.co/datasets/timeblindness/spooky-bench/resolve/main/spooky_bench.zip
unzip spooky_bench.zip
License: MIT License