DeepUSPS: Deep Robust Unsupervised Saliency Prediction With Self-Supervision

Nguyen, Duc Tam; Dax, Maximilian; Mummadi, Chaithanya Kumar; Ngo, Thi Phuong Nhung; Nguyen, Thi; Lou, Zhongyu; Brox, Thomas

doi:10.48550/arxiv.1909.13055

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As the results of this paper demonstrate, self-supervision does not necessarily improve classification accuracy under benchmarking conditions, but the considerable advantage of the method is its robustness towards several potentially problematic data set characteristics as well as the possibility of using data sets for which no labels exist. This is consistent with previous research in this field (Hendrycks et al, 2019;Nguyen et al, 2021;Noroozi & Favaro, 2016). According to our results, self-supervised learning is especially superior to the conventional method when learning is made very difficult due to different kinds of data manipulation, while there is only a slight advantage for minor manipulations.…”

Section: Discussionsupporting

confidence: 92%

Improving image classification robustness using self‐supervision

Wittscher

Diers

Pigorsch

2022

Stat

View full text Add to dashboard Cite

Self‐supervised learning allows training of neural networks without immense, high‐quality or labelled data sets. We demonstrate that self‐supervision furthermore improves robustness of models using small, imbalanced or incomplete data sets which pose severe difficulties to supervised models. For small data sets, the accuracy of our approach is up to 12.5% higher using MNIST and 15.2% using Fashion‐MNIST compared to random initialization. Moreover, self‐supervision influences the way of learning itself, which means that in case of small or strongly imbalanced data sets, it can be prevented that classes are not or insufficiently learned. Even if input data are corrupted and large image regions are missing from the training set, self‐supervision significantly improves classification accuracy (up to 7.3% for MNIST and 2.2% for Fashion‐MNIST). In addition, we analyse combinations of data manipulations and seek to generate a better understanding of how pretext accuracy and downstream accuracy are related. This is not only important to ensure optimal pretraining but also for training with unlabelled data in order to find an appropriate evaluation measure. As such, we make an important contribution to learning with realistic data sets and making machine learning accessible to application areas that require expensive and difficult data collection.

show abstract

Section: Discussionsupporting

confidence: 92%