Deep Image Retrieval: Learning global representations for image search. In ECCV, 2016.
论文地址:https://arxiv.org/abs/1604.01325
extended version:end to end learning of deep visual representations for image retrieval, in arxiv, 16.10.
=====
看图说话:
从图上可以看出该论文的整体框架:
1 基于pre-trained model on Imagenet(如VGG16)
2 从Landmarks dataset[17],中挖掘出一个full或者clean的数据集(包括类别标签 Full Datset和bounding box Clean Dataset)
3 用数据集Full Daset来进行finetune,其loss是一般的分类的lo