Datasetpublic

suijinru/MultiViewBench

MultiView-Bench Evaluation data for MultiView-Bench: A Diagnostic Benchmark for World-Centric Multi-View Integration in VLMs (arXiv:2607.08970). This repository currently contains the synthetic image-and-text portion of the benchmark. The real-world subset is not included; see the scope note below. Contents 2,400 evaluation samples across 24 benchmark variants 7,900 PNG views (2.94 GiB logical image data) One, three, or six images per sample English question text… See the full description on the dataset page: https://huggingface.co/datasets/suijinru/MultiViewBench.

sourceHugging Facemitupdated 2mo agoView on Hugging Face

1likes1kdownloads

26 commits on main

b22b3632mo ago

Update release checksums for task previews

suijinru

d0beb0a2mo ago

Add diverse task previews

suijinru

05210db2mo ago

Update release checksums

suijinru

2c970282mo ago

Rename paper_row to task_variant

suijinru

106c68b2mo ago

Update README.md

suijinru

99954db2mo ago

Set dataset license to MIT

suijinru

3c8e7bf2mo ago

Align release manifest with canonical dataset name

suijinru

a6e9bef2mo ago

Update dataset card and task previews

suijinru

823734e2mo ago