Outfit recommendation often requires a complex and large model that incurs huge memory and time costs . The explosive number of outfit candidates amplifies the data sparsity problem, often leading to poor outfit representation . False NegativeDistillation (FND) exploits false-negative information from the teacher model while not requiring the ranking of all candidates . We introduce a CLframework for outfit representation learning with two proposed dataaugmentation methods. Quantitative and qualitative experiments on outfitrecommendation datasets demonstrate the effectiveness and soundness of our proposed methods. We propose a new KD framework for outfit recommendation, called FND, which exploits false negative information from an apretrained teacher model. It was inspired by the recent success of contrastive learning (CL) to tackle this issue, we introduce a new CL framework for Outfit Recommendation and FND. It is available to download and test your outfit recommendation. For more information, please visit www.fault.com/Outfit/FND .

Author(s) : Seongjae Kim, Jinseok Seol, Holim Lim, Sang-goo Lee

Links : PDF - Abstract

Code :
Coursera

Keywords : outfit - recommendation - false - fnd - negative -

Leave a Reply

Your email address will not be published. Required fields are marked *