README.md

metadata

dataset_info:
  - config_name: fall
    features:
      - name: image
        dtype: image
      - name: embedding
        list: float16
      - name: series_name
        dtype: string
      - name: episode_name
        dtype: string
      - name: frame_index
        dtype: int32
    splits:
      - name: train
        num_bytes: 152222868608
        num_examples: 825046
    download_size: 152070943196
    dataset_size: 152222868608
  - config_name: spring
    features:
      - name: image
        dtype: image
      - name: embedding
        list: float16
      - name: series_name
        dtype: string
      - name: episode_name
        dtype: string
      - name: frame_index
        dtype: int32
    splits:
      - name: train
        num_bytes: 144942749284
        num_examples: 772672
    download_size: 144807609295
    dataset_size: 144942749284
  - config_name: summer
    features:
      - name: image
        dtype: image
      - name: embedding
        list: float16
      - name: series_name
        dtype: string
      - name: episode_name
        dtype: string
      - name: frame_index
        dtype: int32
    splits:
      - name: train
        num_bytes: 125949651243
        num_examples: 689084
    download_size: 125818768936
    dataset_size: 125949651243
  - config_name: winter
    features:
      - name: image
        dtype: image
      - name: embedding
        list: float16
      - name: series_name
        dtype: string
      - name: episode_name
        dtype: string
      - name: frame_index
        dtype: int32
    splits:
      - name: train
        num_bytes: 114073573877
        num_examples: 615698
    download_size: 113960327223
    dataset_size: 114073573877
configs:
  - config_name: fall
    data_files:
      - split: train
        path: fall/train-*
  - config_name: spring
    data_files:
      - split: train
        path: spring/train-*
  - config_name: summer
    data_files:
      - split: train
        path: summer/train-*
  - config_name: winter
    data_files:
      - split: train
        path: winter/train-*
extra_gated_heading: Request Access to Anime Video Dataset 2024 - Frame Embeddings
extra_gated_description: >
  Access to this dataset is restricted to **academic and non-commercial
  research**. Please provide the following information to request access.  
  Submissions will be reviewed manually. Approval is not guaranteed.
extra_gated_prompt: >
  By submitting this form, you confirm that: 1. You will use this dataset
  **solely for non-commercial research purposes**. 2. You will **not
  redistribute, resell, or use** the data for commercial applications. 3. You
  acknowledge that all materials remain the copyright of their respective
  owners. 4. You agree to follow fair use and research ethics standards when
  using the dataset.
extra_gated_fields:
  Full Name: text
  Institution or Organization: text
  Country: country
  Research Area:
    type: select
    options:
      - Multimedia Analysis
      - Audio-Visual Generation
      - Multimodal Retrieval
      - Speech or VAD Studies
      - Other
  Intended Research Description: text
  Estimated Project Duration (months): text
  Expected Research Output:
    type: select
    options:
      - Academic Paper
      - Thesis or Dissertation
      - Open Research Model
      - Internal Report
      - Other
  Planned Start Date: date_picker
  I agree to use this dataset for non-commercial research ONLY: checkbox
extra_gated_button_content: Submit Request

Anime Video Dataset 2024 - Frame Embeddings

Overview

This dataset provides pre-computed visual embeddings for anime video frames from the 2024 anime season. Each frame has been processed using the SigLIP 2 vision model to generate high-quality visual embeddings suitable for various multimodal research tasks. The dataset is derived from JacobLinCool/anime-2024 and includes seasonal subsets (Winter, Spring, Summer, and Fall).

Data Details

• Source Dataset: JacobLinCool/anime-2024
• Embedding Model: google/siglip2-so400m-patch16-naflex
• Frame Extraction Rate: 1 frame per second (from 23.98 FPS source)
• Maximum Duration: First 120 minutes of each episode
• Embedding Dimension: 1152
• Embedding Precision: float16

Schema

Each record in the dataset contains:

• image (Image) — The extracted video frame
• embedding (Sequence[float16]) — The visual embedding vector
• series_name (string) — The title of the anime series
• episode_name (string) — The episode identifier (numeric or special episode designation)
• frame_index (int32) — The frame number within the video (0-indexed)

Acknowledgments

This dataset is derived from the Anime Video Dataset 2024, which sources content from an OTT platform. All videos are provided exclusively for research purposes and are derived from publicly accessible low-resolution content. All original copyrights remain with their respective rights holders.

License

This dataset inherits the license and usage restrictions from the source dataset. It is provided exclusively for research purposes. Commercial use is prohibited.