Frame2KG-YC2 / README.md
Lewis Watson
Align calibration split with special tokens
ef28931
|
Raw
History Blame Contribute Delete
2.97 kB
metadata
dataset_info:
  features:
    - name: video_id
      dtype: string
    - name: frame_number
      dtype: string
    - name: category
      dtype: string
    - name: image
      dtype: image
    - name: graph
      struct:
        - name: edges
          list:
            - name: predicate
              dtype: string
            - name: source
              dtype: string
            - name: target
              dtype: string
        - name: nodes
          list:
            - name: attributes
              struct:
                - name: appearance
                  dtype: string
                - name: size
                  dtype: string
            - name: id
              dtype: string
            - name: label
              dtype: string
            - name: location
              dtype: string
    - name: graph_sentence
      dtype: string
  splits:
    - name: training
      num_bytes: 486523923
      num_examples: 12512
    - name: train_calibration
      num_bytes: 9893243
      num_examples: 256
    - name: validation
      num_bytes: 117826126
      num_examples: 3357
    - name: testing
      num_bytes: 112245657
      num_examples: 2849
    - name: validation_dev
      num_bytes: 3297153
      num_examples: 100
  download_size: 710966092
  dataset_size: 729786102
configs:
  - config_name: default
    data_files:
      - split: training
        path: data/training-*
      - split: train_calibration
        path: data/train_calibration-*
      - split: validation
        path: data/validation-*
      - split: testing
        path: data/testing-*
      - split: validation_dev
        path: data/validation_dev-*

Frame2KG-YC2

Frame2KG-YC2 is a frame-to-knowledge-graph synthetic dataset derived from YouCook2, a cooking video dataset.

Each example contains an image, source metadata, a structured graph of localised entities and relations, and a short graph-derived sentence.

Splits

The training, validation, and testing splits are the primary dataset splits.

The validation_dev split is a 100-example subsample of validation.

The train_calibration split is a 256-example subsample of training, intended for calibration or quantisation workflows.

Citation

If you use this dataset in your work, please cite the paper:

@inproceedings{watson2026frame2kg,
  title = {Frame2KG: A Benchmark and Evaluation Toolkit for Interpretable Frame-to-Graph Generation},
  author = {Watson, Lewis N. and Strathearn, Carl and Mitchell, Kenny and Yu, Yanchao},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  month = {May},
  year = {2026},
  pages = {10912--10926},
  address = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio},
  doi = {10.63317/4ys6kofrzoc5},
  url = {https://doi.org/10.63317/4ys6kofrzoc5}

Notes

Annotations are synthetic and cooking-domain specific, so the dataset should be treated as a controlled benchmark rather than general-purpose visual ground truth. Dataset is provided as is.