Google通用图像嵌入第6名解决方案

Oct, 2022

6th Place Solution to Google Universal Image Embedding

S. Gkelios, A. Kastellos, S. Chatzichristofis

TL;DR基于 CLIP 架构，使用 SubCenter ArcFace loss ,dynamic margins 和提高 transfer learning 策略，以 Google Universal Image Embedding 竞赛为基础，创建了多样化的数据集，最终在 Kaggle 的排名中获得第六名，私有排行榜得分为 0.685。

Abstract

This paper presents the 6th place solution to the google universal image embedding competition on Kaggle. Our approach is based on the clip architecture, a powerful pre-trained model used to learn visual represen