Dataset Viewer (First 5GB)
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 528222474 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

# big_city 完整样本 Arrow 包说明

本目录由 scripts/pack_big_city_hf_arrow.py 生成。打包流程会先做完整性检测,然后只保留完整样本;检测为不完整的样本不会进入 assets/,也不会进入 arrow/files-*.arrow

完整性筛选

  • 原始样本数:128
  • 完整样本数:63
  • 不完整样本数:65
  • 实际打包样本数:63
  • 完整 weather 条目行数:189
  • 完整 frame 行数:9828

本次数据中完整样本为 big_city_random_4m_032_01big_city_random_4m_032_01_v062。从 big_city_random_4m_032_01_v063 开始的样本缺少三种天气下的 RGB/mask/manifest 等文件,因此已被排除。

包内容

  • arrow/scenes.arrow:每个完整体素场景一行。
  • arrow/weather.arrow:每个完整样本的每种天气一行。
  • arrow/frames-*.arrow:每个完整渲染 frame 一行。
  • arrow/files-*.arrow:启用 --embed-all-files 时生成,每个原始文件一行,含 content bytes。
  • assets/:启用 --copy-assets 时生成,只包含完整样本的完整目录树。
  • completeness_summary.json:完整性统计。
  • missing_entries.csv:被排除样本的缺失明细。
  • complete_samples.txt:进入包的完整样本列表。
  • incomplete_samples.txt:被排除的不完整样本列表。

当前资产模式

  • copy_assetsnone
  • asset_scopesample-tree
  • embedded_all_filesTrue
  • assets/ 文件数:0
  • assets/ 逻辑大小:0 bytes
  • arrow/files-*.arrow 文件行数:45048
  • arrow/files-*.arrow 内容大小:148923092194 bytes

本包已经启用 embedded_all_files=true,所有完整样本的原始文件 bytes 都在 arrow/files-*.arrowcontent 字段里。

PyArrow 读取示例

import pyarrow.ipc as ipc

with ipc.open_stream("arrow/scenes.arrow") as reader:
    scenes = reader.read_all()

with ipc.open_stream("arrow/frames-00000.arrow") as reader:
    frames = reader.read_all()

print(scenes.schema)
print(frames.schema)

如果启用了 --embed-all-files,可以这样取回原始文件:

import pyarrow.ipc as ipc
from pathlib import Path

with ipc.open_stream("arrow/files-00000.arrow") as reader:
    files = reader.read_all()

row = files.slice(0, 1).to_pydict()
rel_path = row["rel_path"][0]
content = row["content"][0]

out_path = Path("restored") / rel_path
out_path.parent.mkdir(parents=True, exist_ok=True)
out_path.write_bytes(content)

生成命令

当前 hardlink 完整目录包:

python scripts\pack_big_city_hf_arrow.py --copy-assets hardlink --asset-scope sample-tree --rows-per-frame-shard 4096

纯 Arrow 内嵌全文件包:

python scripts\pack_big_city_hf_arrow.py `
  --embed-all-files `
  --file-shard-bytes 536870912 `
  --rows-per-frame-shard 4096 `
  --output-dir "E:\Dataset\bc32x3_vox52_fullraw256_clean_pts25_128poses\run_20260516_145538\big_city_hf_arrow_embedded"
Downloads last month
569