Implementation:Open compass VLMEvalKit CGBench MCQ Grounding Mini

Field	Value
source	VLMEvalKit
domain	Vision, Benchmarking, Video MCQ, Temporal Grounding

Overview

Benchmark dataset implementation for CG-Bench MCQ with temporal grounding (mini split) evaluation in VLMEvalKit.

Description

CGBench_MCQ_Grounding_Mini inherits from VideoBaseDataset and implements the mini split of the CG-Bench benchmark combining multiple-choice question answering with temporal grounding in long videos. The TYPE field is set to 'Video-MCQ-Grounding'. The file also defines CGBench_OpenEnded_Mini, CGBench_MCQ_Grounding, and CGBench_OpenEnded classes for different evaluation modes and data splits.

Usage

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

Code Reference

Source: vlmeval/dataset/cgbench.py, Lines: L1-1765
Import: from vlmeval.dataset.cgbench import CGBench_MCQ_Grounding_Mini

Signature:

class CGBench_MCQ_Grounding_Mini(VideoBaseDataset):
    TYPE = "Video-MCQ-Grounding"
    dataset = "CG-Bench_MCQ_Grounding_Mini"
    MD5 = "54ed3e90a51a6fb375c92b319a715f72"
    ...

class CGBench_OpenEnded_Mini(VideoBaseDataset):
    TYPE = "Video-OpenEnded"
    ...

class CGBench_MCQ_Grounding(VideoBaseDataset):
    TYPE = "Video-MCQ-Grounding"
    ...

class CGBench_OpenEnded(VideoBaseDataset):
    TYPE = "Video-OpenEnded"
    ...

I/O Contract

Direction	Description
Inputs	TSV dataset file with video paths, MCQ questions, and grounding annotations
Outputs	Evaluation results DataFrame with accuracy and grounding scores

Usage Examples

from vlmeval.dataset import build_dataset
dataset = build_dataset('CG-Bench_MCQ_Grounding_Mini')

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment