Deep convolutional image retrieval: A general framework
作者：
Highlights：
•
• This work is able to exploit any kind of available information about the retrieval task, proposing three retraining approaches: The Fully Unsupervised, the Retraining with Relevant Information, and the Relevance Feedback based retraining.
• Combinatory schemes can be deployed, where all the above approaches can be employed in a pipeline.
• A query expansion technique with a spatial verification step is proposed.
• This work uses retargeting for the learning phase, instead of triplet loss, allowing for single sample training which is very fast.
摘要
•This work is able to exploit any kind of available information about the retrieval task, proposing three retraining approaches: The Fully Unsupervised, the Retraining with Relevant Information, and the Relevance Feedback based retraining.•Combinatory schemes can be deployed, where all the above approaches can be employed in a pipeline.•A query expansion technique with a spatial verification step is proposed.•This work uses retargeting for the learning phase, instead of triplet loss, allowing for single sample training which is very fast.
论文关键词：Content based image retrieval,Convolutional neural networks,Deep learning,Query expansion
论文评审过程：Received 23 August 2017, Revised 12 December 2017, Accepted 25 January 2018, Available online 2 February 2018, Version of Record 7 February 2018.
论文官网地址：https://doi.org/10.1016/j.image.2018.01.007
