Implementation:Open compass VLMEvalKit VSIBench

Field	Value
source	VLMEvalKit
domain	Vision, Benchmarking, Video Spatial Intelligence

Overview

Benchmark dataset implementation for VSIBench evaluation in VLMEvalKit.

Description

VSIBench inherits from VideoBaseDataset and implements the VSIBench benchmark for video spatial intelligence evaluation. The TYPE field is set to 'VIDEO-MCQ-and-NA'. It combines multiple-choice (direction, distance, route planning) and numerical answer (counting, size estimation) question types.

Usage

Registered in vlmeval/dataset/__init__.py and invoked through build_dataset() by benchmark name.

Code Reference

Source: vlmeval/dataset/vsibench.py, Lines: L1-227
Import: from vlmeval.dataset.vsibench import VSIBench

Signature:

class VSIBench(VideoBaseDataset):
    TYPE = 'VIDEO-MCQ-and-NA'
    DATASET_URL = {...}
    DATASET_MD5 = {...}
    ...

I/O Contract

Direction	Description
Inputs	TSV dataset file with image/video paths and questions
Outputs	Evaluation results DataFrame with scores per category

Usage Examples

from vlmeval.dataset import build_dataset
dataset = build_dataset('VSIBench')

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment