SOCO-LVLM

SOCO-LVLM provides multiple-choice semantic object correspondence evaluation data for LVLMs. This is the SOCO-LVLM v1 release, derived from SOCOv1. The original SOCO correspondence benchmark is available in the GenIntelLab/SOCO dataset repository.

Repository Layout

GenIntelLab/SOCO-LVLM
  SOCO_LVLM/
    soco_lvlm_img.tsv
    soco_lvlm_imgtxt.tsv
    soco_lvlm_txt.tsv
  README.md

Variants

• soco_lvlm_img.tsv: image-input evaluation variant (approximately 3.24 GB).
• soco_lvlm_imgtxt.tsv: image-and-text evaluation variant (approximately 3.24 GB).
• soco_lvlm_txt.tsv: text-input evaluation variant (approximately 1.63 GB).

Each TSV uses the columns question, image, image_path, answer, index, g_index, qid, category, A, B, C, and D.

Download

Install the Hub client:

pip install -U huggingface_hub

Download all three variants:

hf download GenIntelLab/SOCO-LVLM --repo-type dataset --local-dir SOCO-LVLM

Download only one variant in Python:

from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="GenIntelLab/SOCO-LVLM",
    repo_type="dataset",
    filename="SOCO_LVLM/soco_lvlm_img.tsv",
)

Replace the filename with soco_lvlm_imgtxt.tsv or soco_lvlm_txt.tsv to select a different evaluation variant.

Citation

@misc{duenkel2026soco,
  title         = {SOCO: Benchmarking Semantic Object Correspondence in Vision Foundation Models},
  author        = {D{\"u}nkel, Olaf and Sunagad, Basavaraj and Wang, Haoran and
                   Hoffmann, David T. and Theobalt, Christian and Kortylewski, Adam},
  year          = {2026},
  eprint        = {2605.31597},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2605.31597}
}