中文文档 |英语

一个基于Python的工具包，用于构建昆虫图像数据集。EntomoKit提供了 用于帧提取、分割和形态的统一entomokit CLI处理 测量、综合、清理、增强、数据集拆分、AutoMM 分类和环境诊断，以及entomokit-workflow技能 人工智能助理。

工作流程概述

流水线是可组合的;只需完成你数据所需的步骤。

1. extract-frames — 将视频转换为静态画面。
2. segment — 从图像中剪取昆虫，并可选地写注释。
3. measure — 可选：来自分段掩码的形态度量。
4. synthesize — 可选：在新背景上添加复合RGBA剪影。
5. clean — 调整大小、填充和去重图像。
6. augment — 可选：扩展一个小型训练集。
7. split-csv — 构建训练/评估/测试CSV。
8. classify train → classify predict / classify evaluate / classify embed / classify cam / classify export-onnx。

你可以从任意步骤开始：例如直接从clean到classify train 当图像已经被标记时，或者仅使用 segment 进行掩膜提取。

Insect Dataset Toolkit (EntomoKit)

中文文档 | English

A Python-based toolkit for building insect image datasets. EntomoKit provides a unified entomokit CLI for frame extraction, segmentation, morphology measurement, synthesis, cleaning, augmentation, dataset splitting, AutoMM classification and environment diagnostics, plus an entomokit-workflow skill for AI assistants.

Workflow Overview

The pipeline is composable; only the steps your data needs are required.

1. extract-frames — turn videos into still frames.
2. segment — cut insects out of images and optionally write annotations.
3. measure — optional: morphology metrics from segmentation masks.
4. synthesize — optional: composite RGBA cutouts onto new backgrounds.
5. clean — resize, pad and deduplicate images.
6. augment — optional: expand a small training set.
7. split-csv — build train/val/test CSVs.
8. classify train → classify predict / classify evaluate / classify embed / classify cam / classify export-onnx.

You can start at any step: for example go straight from clean to classify train when the images are already labelled, or use segment alone for mask extraction. AI-assisted runs follow the skill’s own guided policy, not this list.

Requirements and Installation

• Python 3.9+
• Linux, macOS or Windows

git clone https://github.com/xtmtd/entomokit.git
cd entomokit

Install into an isolated environment (recommended):

uv venv .venv
source .venv/bin/activate
uv pip install -e .

Equivalent alternatives: python -m venv .venv && source .venv/bin/activate, or conda create -n entomokit python=3.11 -y && conda activate entomokit. Installing directly into a global environment is possible but risks dependency conflicts.

Command extras:

measure and synthesize also import scikit-image, which neither the measurement nor the synthesis extra declares; install .[segmentation] (which declares it) or add scikit-image explicitly alongside those extras.

Install several at once, for example the full development set:

uv pip install --only-binary :all: stringzilla
pip install -e ".[dev,classify,segmentation,synthesis,measurement,video,cleaning,augment]"

The stringzilla line above is required before installing .[augment] on platforms where only its binary wheel is available. AutoMM publishes its own install notes at