界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

setup-and-configuration

setup-and-configuration

【效率与协作】Set up FastReID from a source-only checkout, verify imports, and manage config creation, merge, freeze, YAML inheritance, command-line overrides, and model-zoo recipe selection.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: setup-and-configuration description: "Set up FastReID from a source-only checkout, verify imports, and manage config creation, merge, freeze, YAML inheritance, command-line overrides, and model-zoo recipe selection." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Setup and Configuration

Use this sub-skill for FastReID environment preparation and config inspection. It covers source-only install notes, import checks, config creation and merge, _BASE_ inheritance, command-line opts, freeze handling, and model-zoo recipe selection.

Use this sub-skill when

  • You need to confirm that a FastReID checkout imports cleanly from source.
  • You need to merge a config, apply opts, freeze it, or inspect selected keys.
  • You need to choose the right recipe/config family for a benchmark or model zoo entry.

Route elsewhere when needed

  • Dataset directory layouts, dataset registration, and loader validation belong to the dataset sub-skill.
  • Training/evaluation launch, distributed flags, and checkpoint workflows belong to the training sub-skill.
  • Backbone/head/loss/model registry details belong to the modeling sub-skill.
  • ONNX, Caffe, TensorRT, and project-extension deployment flows belong to the deployment sub-skill.

Bundled references

Bundled script

  • scripts/config_merge_check.py — merge a config with optional --opts, print selected keys, and optionally freeze the result without training or downloads.

Public facts to remember

  • FastReID is version 1.3.
  • The checkout is source-only, so import help comes from a repo-root sys.path entry or a private .pth file, not from editable distribution metadata.
  • get_cfg() returns a CfgNode.
  • merge_from_file() supports _BASE_ and defaults to safe YAML loading.
  • merge_from_list() applies CLI overrides.
  • MODEL.DEVICE defaults to cuda, so CPU dry-runs must override it explicitly.
我的收藏