📄 中文摘要
在产品中，图像识别如果不是系统的核心功能，而仅仅是一个特性，后端可能会承担不必要的工作。直接将模型集成到后端并不是问题的根源，真正的问题在于模型的复杂性应当如何管理，以避免应用程序内部逐渐变得混乱。当AI是产品的核心时，可以围绕其构建系统；但当AI仅是更大系统中的一个功能时，如何处理模型逻辑就显得尤为重要。保持系统的整洁与高效是设计时需要考虑的关键因素。
一种不将模型逻辑硬编码到后端的 AI 图像识别生产模式
📄 English Summary
A Production Pattern for AI Image Recognition Without Hardwiring Model Logic Into Your Backend
Integrating image recognition directly into the backend of a product can lead to complications when this feature is not the main focus of the system. The core issue lies not in the use of SDKs but in managing the complexity of the model to prevent the application from deteriorating over time. When AI serves as the product, it is justifiable to build the system around it. However, when AI is merely one function within a larger system, careful consideration is needed to ensure that the model logic is handled appropriately, maintaining the overall integrity and efficiency of the application.