3D1: separately packaged basic and fancy cases
The two independent folders contain all 2,616 current human-approved cases and 2,976 questions. They have disjoint case, question and media sets. Failed or uncertain cases, historical versions, private gold answers, review notes and private source evidence are excluded.
| Folder / configuration | Cases | Questions | Videos | Images |
|---|---|---|---|---|
| basic — 基础 | 1,892 | 2,252 | 2,078 | 99 |
| fancy | 724 | 724 | 724 | 0 |
fancy is identified by the original fancy source batches. basic contains the remaining approved cases: 1,882 main-library cases plus eight diversity and two catalog-40 cases. These are source groups, not new difficulty judgments.
Each folder has its own data/, media/, viewer/, evaluation/scorer.py, schema.json, manifest.json and README. There is no combined default table or shared media directory.
Load a subset
from datasets import load_dataset
basic = load_dataset("flymanss/3D1", "basic", split="unassigned")
fancy = load_dataset("flymanss/3D1", "fancy", split="unassigned")
Authenticate with an account permitted to access the repository if Hugging Face requires access. For local use, load basic/data/questions.parquet or fancy/data/questions.parquet with the Parquet loader.
Media paths in a table are relative to its subset folder. For example, a basic row's media/abc.mp4 resolves to basic/media/abc.mp4 in the repository. The tables store media records and paths, not Hugging Face Video features. Each subset is independently downloadable with snapshot_download(..., allow_patterns=["basic/**"]) or allow_patterns=["fancy/**"].
Questions and evaluation
Both tables retain the same 18-field schema and original bilingual questions, options and response schemas. English previews display questions, media and response formats without category headings or answers. Open basic/viewer/index.html or fancy/viewer/index.html after downloading the corresponding folder.
Responses must use the strict {"value": ...} JSON wrapper. Original option IDs are preserved, including historical non-English IDs. Gold answers are not included; the scoring code requires the separately held private gold.
| Subset | Boolean | Single choice | Structured |
|---|---|---|---|
| Basic | 1,073 | 477 | 702 |
| Fancy | 638 | 66 | 20 |
Validation and source notes
The split preserves every original row and media byte. Case IDs, sample IDs and media are disjoint between folders; their union exactly matches the approved source snapshot. JSONL, Parquet, preview and per-sample schemas are checked for consistency. Schema version 1.0 and scorer version 2.0 are preserved, and all rows remain split=unassigned. These folders are source subsets, not train/test partitions: two original split-group IDs occur in both subsets and are preserved for downstream leakage-aware splitting.
Missing geometry levels for 667 source cases were previously filled using an explicitly confirmed mapping; existing levels were preserved. Four preserved fancy question-wording issues are documented in fancy/README.md. No new license is assigned.
- Downloads last month
- -