They proposed an Attention-based evidence visualization method, L2-CAF. The method is based on a filter that eliminates all features other than the target class from the feature map, and it has the advantage of being able to handle multiple modes, although it only provides a single-mode signal from Softmax output. It can be implemented without any restriction of the base network structure.
TL;DR
They proposed an Attention-based evidence visualization method, L2-CAF. The method is based on a filter that eliminates all features other than the target class from the feature map, and it has the advantage of being able to handle multiple modes, although it only provides a single-mode signal from Softmax output. It can be implemented without any restriction of the base network structure.
Why it matters:
Paper URL
https://arxiv.org/abs/2007.09748
Submission Dates(yyyy/mm/dd)
2020/07/19
Authors and institutions
Ahmed Taha, Xitong Yang, Abhinav Shrivastava, and Larry Davis
Methods
Results
Comments