Implementation:Open compass VLMEvalKit RefCOCODataset

Field	Value
source	VLMEvalKit
domain	Vision, Benchmarking, Visual Grounding

Overview

Benchmark dataset implementation for RefCOCO visual grounding evaluation in VLMEvalKit.

Description

RefCOCODataset inherits from ImageBaseDataset and implements the RefCOCO benchmark family for visual grounding tasks. The TYPE field is set to 'GROUNDING'. It evaluates models on referring expression comprehension across multiple splits (RefCOCO_val, RefCOCO_testA, RefCOCO_testB, RefCOCO+, RefCOCOg) using IoU threshold-based evaluation with a default threshold of 0.5.

Usage

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

Code Reference

Source: vlmeval/dataset/refcoco.py, Lines: L1-568
Import: from vlmeval.dataset.refcoco import RefCOCODataset

Signature:

class RefCOCODataset(ImageBaseDataset):
    TYPE = 'GROUNDING'
    MODALITY = 'IMAGE'
    IOU_THRESHOLD = 0.5
    DATASET_URL = {...}
    DATASET_MD5 = {...}
    ...

I/O Contract

Direction	Description
Inputs	TSV dataset file with images and referring expressions
Outputs	Evaluation results DataFrame with IoU-based grounding accuracy per split

Usage Examples

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

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment