This research focuses on designing a cost-effective content moderation system for user-generated multimedia content by integrating OpenAI's CLIP model into a FastAPI-based asynchronous RESTful API. Our system primarily targets image content classification, specifically nudity, gore, offensive content, and facial attributes. It seeks to address the challenges of efficient data handling, model inference, and response formatting while maintaining flexibility across different classification scenarios. This study aims to contribute to the field of content moderation by enhancing the accuracy and efficiency of harmful content detection, fostering a safer online environment.
This thesis aims to develop a FastAPI-based web application that effectively integrates the CLIP model for image classification. The specific objectives are:
- To create a FastAPI service that accepts image data either as a file upload or a URL.
- To use the CLIP model efficiently for image classification based on user-specified classes.
- To design the system to dynamically handle multiple classification scenarios and adapt to different classifiers and class inputs.
- To evaluate the system's effectiveness and applicability in various image classification scenarios.
The main goal of this study is to create an affordable, efficient, and adaptable content moderation system for user-generated multimedia content. The proposed system utilizes OpenAI's CLIP model for image classification tasks and FastAPI for designing an asynchronous RESTful API. It is aimed at dynamically handling multiple classification scenarios,thereby enhancing the accessibility and utility of image classification technology in real-world applications.
Our FastAPI and CLIP-based image classifier development remains a work in progress. By maintaining a focus on continuous improvement and adapting to the ever-evolving technological landscape, the application promises to remain a formidable tool for image classification tasks. The enhancements proposed in this paper outlines potential for further refining the application's functionality and utility.