Optimization and application of automatic labeling algorithm for large models in computer vision field

Résumé fourni par la source

Traditional manual labeling methods are time-consuming and error-prone, so the optimization of automatic labeling algorithm becomes the key. In this paper, a multi-modal active learning framework (TMMAL) based on Transformer is proposed. Through the strategies of multi-modal feature fusion, active learning and semi-supervised optimization, the accuracy and efficiency of automatic labeling algorithm are significantly improved. The experimental results show that the performance of TMMAL framework has surpassed that of the traditional model using 100% data when only 70% data is used, and the labeling accuracy has been significantly improved, and it also performs well in processing speed and generalization ability. In addition, this paper also discusses the application cases of the algorithm in intelligent security, medical image analysis and automatic driving, and verifies its effectiveness and practicability in actual scenes.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.