hf_maloccu / README.md
dsaint31's picture
Upload README.md with huggingface_hub
b93270f verified
metadata
configs:
  - config_name: default
    default: true
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: single_image
    data_files:
      - split: train
        path: single_image/train-*
      - split: validation
        path: single_image/validation-*
      - split: test
        path: single_image/test-*

Malocclusal Occlusal Image Dataset

부정교합 교합 이미지 데이터셋

Internal Use Only

This dataset repository is private and intended solely for internal research and development. Redistribution or external use is prohibited.
본 데이터셋 저장소는 비공개(private)이며 내부 연구 및 개발 목적으로만 사용됩니다. 외부 배포 또는 재사용은 금지됩니다.


Overview

This repository provides occlusal image datasets derived from patient-level records. All dataset splits are performed at the patient level to prevent information leakage.
본 저장소는 환자 단위(patient-level) 기록으로부터 생성된 교합 이미지 데이터셋을 제공합니다. 모든 데이터 분할은 환자 단위로 수행되어 정보 누출을 방지합니다.

The dataset is distributed in two configurations within the same Hugging Face dataset repository.
본 데이터셋은 하나의 Hugging Face dataset repository 내에서 두 가지 구성(configuration) 으로 제공됩니다.

Both configurations are always generated and uploaded together by the current build script.
현재 빌드 스크립트에서는 두 구성 모두가 항상 함께 생성 및 업로드됩니다.


Dataset Configurations

데이터셋 구성

default (pairwise)

  • One sample contains a pair of images: right and left occlusal views.
    하나의 샘플은 우측과 좌측 교합 이미지 한 쌍을 포함합니다.
  • Intended for models that jointly process bilateral occlusal information.
    양측 교합 정보를 함께 처리하는 모델을 대상으로 합니다.

single_image

  • One sample contains a single image (either right or left).
    하나의 샘플은 우측 또는 좌측 중 하나의 이미지만 포함합니다.
  • Generated by exploding the pairwise dataset.
    pairwise 데이터셋을 분해(explode)하여 생성됩니다.
  • Intended for single-view or image-only models.
    단일 뷰 또는 이미지 단독 입력 모델을 대상으로 합니다.

Splits

All configurations provide the same split structure:
모든 구성은 동일한 데이터 분할 구조를 가집니다.

  • train
  • validation
  • test

Patient-wise split constraint

  • All samples from the same patient_id always belong to the same split.
    동일한 patient_id에 속한 모든 샘플은 항상 동일한 split에 속합니다.
  • This constraint is enforced identically for both configurations.
    이 제약은 두 구성 모두에 동일하게 적용됩니다.
  • This prevents train–test leakage across configurations.
    이를 통해 구성 간 train–test 정보 누출을 방지합니다.

Dataset Schema

Config: default (pairwise)

Field Type Description
image_r Image Right occlusal image / 우측 교합 이미지
image_l Image Left occlusal image / 좌측 교합 이미지
class_r ClassLabel (0..2) Angle class (right) / 우측 Angle 분류
class_l ClassLabel (0..2) Angle class (left) / 좌측 Angle 분류
dist_r float32 Continuous distance (right) / 우측 연속 거리값
dist_l float32 Continuous distance (left) / 좌측 연속 거리값
patient_id string Patient identifier / 환자 식별자
tx_phase string Treatment phase / 치료 단계
row_id int32 Pairwise row index / pairwise 행 인덱스
idx int32 Source index (tracking) / 원본 인덱스

Config: single_image

Field Type Description
image Image Occlusal image / 교합 이미지
side string "r" or "l" / 우측 또는 좌측
label ClassLabel (0..2) Angle class / Angle 분류
dist float32 Continuous distance / 연속 거리값
patient_id string Patient identifier / 환자 식별자
tx_phase string Treatment phase / 치료 단계
parent_row_id int32 Source pairwise row / 원본 pairwise 행
parent_idx int32 Source index (tracking) / 원본 인덱스

Label Convention (라벨 규칙)

  • Angle class labels use 0-based indexing: {0, 1, 2}
    Angle 분류 라벨은 0부터 시작하는 인덱스 {0, 1, 2}를 사용합니다.
    • 0: Angle Class 1
    • 1: Angle Class 2
    • 2: Angle Class 3
  • Labels are converted once during dataset construction.
    라벨 변환은 데이터셋 생성 시 한 번만 수행됩니다.
  • Class semantics follow the standard Angle classification.
    분류 의미는 표준 Angle 분류 체계를 따릅니다.

Usage (사용 방법)

Load pairwise dataset (default)

from datasets import load_dataset

ds = load_dataset("dsaint31/hf_maloccu")

Load single-image dataset

from datasets import load_dataset

ds_img = load_dataset("dsaint31/hf_maloccu", "single_image")

Notes (참고 사항)

  • Dataset splits are patient-wise, not sample-wise. 데이터 분할은 샘플 단위가 아닌 환자 단위로 수행됩니다.
  • Both configurations always share the same patient split. 두 구성은 항상 동일한 환자 분할을 공유합니다.
  • Tracking fields (row_id, idx, parent_row_id, parent_idx) are provided for internal analysis and traceability. 추적용 필드(row_id, idx, parent_row_id, parent_idx)는 내부 분석 및 정합성 검증을 위해 제공됩니다.
  • Images are stored as grayscale PNGs and embedded into parquet files during Hub upload (embed_external_files=True). 이미지는 grayscale PNG로 저장되며 Hub 업로드 시 parquet 파일에 직접 포함됩니다(embed_external_files=True).

License

Proprietary – Internal use only.

This dataset and any derived artifacts are intended for internal use only. Unauthorized copying, distribution, or modification is prohibited. 본 데이터셋 및 파생 결과물은 내부 사용만을 목적으로 하며, 무단 복제, 배포, 수정은 금지됩니다.