What is it about?
To many researchers CNN is black-box. They use it for different purposes. To understand how does it work, which parts of an image is more important to CNN models, we used different kind of visualization techniques. In context of object detection, we visualized intermediate layers of CNN to understand what are the features it extracts in different layers. We also used Grad-CAM technique to understand which part of image corresponding to a query object.
Featured Image
Photo by David Gabrić on Unsplash
Why is it important?
This work confirms our idea to analyze why CNN fail to detect an object or in context of detection which part of objects matter a most. This work can be used a pre-processing step before developing a new object detection model
Perspectives
This article can help new researchers as well as people working using deep learning models to understand how does a CNN model work in easiest way. They can also understand what type of features are extracted by the model, how much it is important for a model in learning process.
Mr Abhishek Mukhopadhyay
Indian Institute of Science
Read the Original
This page is a summary of: Decoding CNN based Object Classifier Using Visualization, September 2020, ACM (Association for Computing Machinery),
DOI: 10.1145/3409251.3411721.
You can read the full text:
Resources
Contributors
The following have contributed to this page
