README.md

metadata

license: cc-by-4.0
pretty_name: SOCOv1
viewer: false
task_categories:
  - image-to-image
tags:
  - computer-vision
  - keypoint-detection
  - object-correspondence
  - synthetic-data
size_categories:
  - 10K<n<100K

SOCOv1

SOCOv1 is a structured object correspondence dataset with rendered object images, per-view keypoint annotations, pair files, and metadata.

The dataset is distributed as an unpacked folder tree on the Hugging Face Hub. Use huggingface_hub.snapshot_download to download the full dataset or selected subfolders.

Folder Structure

SOCOv1/
  Images/
    <category>/
      *.JPEG
  KeypointAnnotations/
    <category>/
      *.json
  PairAnnotations/
    intra/
      <category>/
        *.json
    cross/
      <category>/
        *.json
    trainsplits/
      train/
        <category>/
          *.json
      test/
        <category>/
          *.json
  Metadata/
    filename_mapping.json
    keypoint_taxonomy.json

Contents

• Images/: rendered object images, organized by category.
• KeypointAnnotations/: per-view keypoint annotations, organized by category.
• PairAnnotations/intra/: intra-category image-pair files.
• PairAnnotations/cross/: cross-category image-pair files.
• PairAnnotations/trainsplits/: train/test split pair files.
• Metadata/: keypoint taxonomy and filename mapping.

This release contains 100 categories, 4,000 images, 4,000 keypoint annotation files, and 60,001 pair annotation files.

Download

Install the Hub client:

pip install -U huggingface_hub

Download the full dataset:

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="GenIntelLab/SOCO",
    repo_type="dataset",
    local_dir="SOCOv1",
    token=True,  # required while the dataset is private
)

Download only selected folders:

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="GenIntelLab/SOCO",
    repo_type="dataset",
    local_dir="SOCOv1_images_and_metadata",
    allow_patterns=["Images/**", "Metadata/**"],
    token=True,
)

Citation

Citation information will be added with the public release.