Dense open-set recognition based on training with noisy negative images

作者：

Highlights：

• Training with noisy negative images greatly improves dense open-set recognition.

• Training with randomly pasted negatives improves generalization on mixed-content images.

• Shared features improve outlier detection and allow for inference with a single forward pass.

• State-of-the-art results on dense open-set recognition benchmarks: WildDash 1, Fishyscapes and StreetHazard.

摘要

•Training with noisy negative images greatly improves dense open-set recognition.•Training with randomly pasted negatives improves generalization on mixed-content images.•Shared features improve outlier detection and allow for inference with a single forward pass.•State-of-the-art results on dense open-set recognition benchmarks: WildDash 1, Fishyscapes and StreetHazard.

论文关键词：Dense prediction,Semantic segmentation,Dense open-set recognition,Outlier detection

论文评审过程：Received 22 January 2021, Revised 31 May 2021, Accepted 24 May 2022, Available online 1 June 2022, Version of Record 11 June 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104490