Capsule Interpretability in Object Detection

Great progress has been made in the field of target detection in recent years. Most of these improvements come from the use of more complex convolutional neural networks. However, how to introduction explanations for individual classifications of features has not been well explored. To overcome the lack of interpretability, we can either propose capsule to have build-in explanations. In this work, we propose interpretability of capsule to replace the original feature extraction. In the proposed model, feature explanations can be created effectively and efficiently. Our model also demonstrates some benefits, experimental results verify the interpretability of the learned features which can help us understand what is the important factors in object detection.