← 返回 Skills 市场
图片向量嵌入技能
作者
Venwell Chiang
· GitHub ↗
· v1.0.0
· MIT-0
116
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install image-embedding
功能描述
图片向量嵌入技能，支持将病害图片转换为特征向量，用于后续的图片检索和相似度匹配。适用于病害知识库的图片特征提取、向量入库等场景。
使用说明 (SKILL.md)
图片向量嵌入技能
核心功能
- 支持JPG/PNG/WebP等常见病害图片格式的特征提取
- 输出1024维归一化特征向量，适配病害检索场景精度要求
- 支持批量处理图片，单批最大支持100张
- 内置预训练病害特征提取模型，无需额外微调即可使用
依赖安装
pip install torch>=2.0 torchvision>=0.15 transformers>=4.30 pillow>=10.0
使用方式
单张图片提取
from scripts.embedding import ImageEmbedding
embeder = ImageEmbedding()
feature_vector = embeder.extract("./test_disease.jpg")
批量提取
vectors = embeder.batch_extract(["./img1.jpg", "./img2.jpg"])
输出格式
返回归一化后的numpy数组，shape为(1024,)
安全使用建议
This skill's documentation promises a real pretrained image embedding model and lists heavy ML dependencies, but the included code is a placeholder that returns random vectors. Before installing or using it: (1) don't assume outputs are meaningful — test with known inputs to verify vectors (e.g., check reproducibility, normalization, similarity behavior); (2) avoid blindly running 'pip install' of large ML libraries unless you actually need them — the shipped code only needs numpy; (3) ask the author for the real model implementation or for SKILL.md to be corrected (provide model weights, license, inference code, and how normalization/batch limits are enforced); (4) if you need production embeddings, prefer a skill with explicit, auditable model code or a link to a trusted model/release. If the author supplies a true model implementation that matches the docs, re-evaluate; otherwise treat this as untrusted/testing-only code.
功能分析
Type: OpenClaw Skill
Name: image-embedding
Version: 1.0.0
The skill provides a placeholder implementation for image feature extraction (embedding) for disease-related images. The code in scripts/embedding.py is a simple mock that returns random vectors, and the SKILL.md documentation describes standard usage of common machine learning libraries (torch, torchvision, transformers) without any malicious instructions or suspicious behaviors.
能力评估
Purpose & Capability
SKILL.md and description claim a pre-trained disease-specific embedding model, 1024-d normalized outputs, support for common image formats, batching limits, and ML dependencies (torch, torchvision, transformers, pillow). The included code (scripts/embedding.py) contains only a simple class that returns random numpy vectors and does not use any of the declared ML libraries or implement format handling, normalization, or batch-size enforcement. The requested dependencies and promised capabilities are disproportionate to the actual implementation.
Instruction Scope
Runtime instructions tell the user/agent to install heavy ML packages via pip, and show usage examples that imply a real model. The SKILL.md instructs operations that the code does not implement (pretrained model extraction, normalization, file format handling, max-batch enforcement). There are no instructions to access unrelated files or credentials, but the guidance is misleading about runtime behavior.
Install Mechanism
There is no formal install spec (skill is instruction-only), but SKILL.md recommends pip installing torch, torchvision, transformers, pillow. Because those installs are not reflected in code and may be unnecessary, this is an inconsistency (not a direct supply-chain risk here, but installing large ML packages without need is avoidable risk).
Credentials
The skill requests no environment variables, credentials, or config paths. There is no sign of unnecessary credential access or exfiltration in code or instructions.
Persistence & Privilege
Skill flags are default (always: false, user-invocable: true, model invocation allowed). The skill does not request elevated persistence or modify other skills; no concern here.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install image-embedding - 安装完成后，直接呼叫该 Skill 的名称或使用
/image-embedding触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
Initial release of image-embedding skill:
- Converts disease images (JPG/PNG/WebP) into 1024-dimensional normalized feature vectors.
- Enables both single and batch image processing (up to 100 images per batch).
- Integrates a pre-trained disease feature extraction model for out-of-the-box use.
- Designed for image retrieval and similarity matching in disease knowledge bases.
元数据
常见问题
图片向量嵌入技能 是什么？
图片向量嵌入技能，支持将病害图片转换为特征向量，用于后续的图片检索和相似度匹配。适用于病害知识库的图片特征提取、向量入库等场景。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 116 次。
如何安装 图片向量嵌入技能？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install image-embedding」即可一键安装，无需额外配置。
图片向量嵌入技能 是免费的吗？
是的，图片向量嵌入技能 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
图片向量嵌入技能 支持哪些平台？
图片向量嵌入技能 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 图片向量嵌入技能？
由 Venwell Chiang（@kumamon2019s）开发并维护，当前版本 v1.0.0。
推荐 Skills