$n$-Reference Transfer Learning for Saliency Prediction

9 Jul 2020  ยท  Yan Luo, Yongkang Wong, Mohan S. Kankanhalli, Qi Zhao ยท

Benefiting from deep learning research and large-scale datasets, saliency prediction has achieved significant success in the past decade. However, it still remains challenging to predict saliency maps on images in new domains that lack sufficient data for data-hungry models. To solve this problem, we propose a few-shot transfer learning paradigm for saliency prediction, which enables efficient transfer of knowledge learned from the existing large-scale saliency datasets to a target domain with limited labeled examples. Specifically, very few target domain examples are used as the reference to train a model with a source domain dataset such that the training process can converge to a local minimum in favor of the target domain. Then, the learned model is further fine-tuned with the reference. The proposed framework is gradient-based and model-agnostic. We conduct comprehensive experiments and ablation study on various source domain and target domain pairs. The results show that the proposed framework achieves a significant performance improvement. The code is publicly available at \url{https://github.com/luoyan407/n-reference}.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Transfer Learning for Saliency Prediction SALICON->WebpageSaliency - 10-shot DINet+FT|Ref NSS 1.6439 # 1
AUC 0.8276 # 1
CC 0.6605 # 1
Few-Shot Transfer Learning for Saliency Prediction SALICON->WebpageSaliency - 1-shot DINet+FT|Ref NSS 1.5077 # 1
AUC 0.8051 # 1
CC 0.6121 # 1
Few-Shot Transfer Learning for Saliency Prediction SALICON->WebpageSaliency - 1-shot ResNet+FT|Ref NSS 1.4272 # 2
AUC 0.7983 # 2
CC 0.5817 # 2
Few-Shot Transfer Learning for Saliency Prediction SALICON->WebpageSaliency - 5-shot DINet+FT|Ref NSS 1.6085 # 1
AUC 0.8200 # 1
CC 0.6468 # 1
Few-Shot Transfer Learning for Saliency Prediction SALICON->WebpageSaliency - EUB DINet+FT|Ref NSS 1.8831 # 1
AUC 0.8494 # 1
CC 0.7442 # 1

Methods


No methods listed for this paper. Add relevant methods here