AGENT SKILL

img2dataset

Create, restart, audit, and scale img2dataset runs for turning URL tables into image datasets.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
img2dataset
license
MIT
metadata
{ "disco-role": "operating" }
description
Create, restart, audit, and scale img2dataset runs for turning URL tables into image datasets.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

img2dataset

Use this skill when a user wants to turn a URL list or table into an image dataset with img2dataset, or needs to debug an existing download run.

What this skill covers

  • The public Python API img2dataset.download(...) and the img2dataset console command.
  • Input table formats and output writer layouts.
  • Image resizing, encoding, and bounding-box blurring.
  • Distributed execution with multiprocessing, PySpark, and Ray.
  • Restarting interrupted runs, hash verification, SSL policy, and X-Robots-Tag handling.
  • Dataset-scale recipes and common troubleshooting.

Start here

  1. Confirm the package is installed and reachable:

    python -c "from importlib.metadata import version; from img2dataset import download; print(version('img2dataset')); print(download)"
    
  2. Install the package in the environment you want to inspect. For a release install, use python -m pip install img2dataset; for a local checkout, use python -m pip install -e . from the checkout root.

  3. Use the sub-skill that matches the user's request:

    • core-download for CLI/API runs, retries, incremental recovery, hashes, SSL, and X-Robots-Tag policy.
    • input-output-formats for input schemas, output formats, metadata layouts, captions, hashes, and TFRecord prerequisites.
    • image-processing for resize modes, codecs, filters, and bbox blur.
    • distributed-execution for multiprocessing, PySpark, Ray, W&B, throughput, and cluster troubleshooting.

Core import and CLI check

When you only need to confirm the package is alive, run one of these from a neutral working directory:

python -c "from img2dataset import download; print(download)"
img2dataset -- --help

If img2dataset is not on PATH, use the CLI help check described in core-download.

Read before relying on this skill

  • Repository provenance if you need to confirm whether this skill matches the current checkout.
  • Dataset recipes for common public dataset command patterns and scale notes.
  • Troubleshooting for install/import failures, backend dependencies, filesystem prefixes, and stale-skill checks.
  • Repo routing metadata is consumed by the managed repo-skill router during import.

Root helper

  • scripts/check_img2dataset_env.py checks the installed package, signature, CLI help, and optional backend availability without depending on the original checkout.

Route map

  • Use core-download when the request is about starting, restarting, or recovering a run; deciding incremental, overwrite, or extend; or handling hashes, SSL, robots directives, and base metadata.
  • Use input-output-formats when the request is about --input_format, --output_format, output layout, metadata schema, or TFRecord and parquet sidecars.
  • Use image-processing when the request is about Resizer, ResizeMode, skip_reencode, disable_all_reencoding, interpolation, image size filters, or bbox blur.
  • Use distributed-execution when the request is about --distributor, PySpark, Ray, W&B, process/thread tuning, subjob_size, or throughput bottlenecks.

Notes for future agents

  • Keep runtime guidance inside this skill tree; do not point back to the original repository checkout.
  • Treat dataset-scale public recipes as reference material, not as default verification cases.
  • If the current checkout has changed, compare it with references/repo-provenance.md before reusing the skill.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351