[Paper Review] CapsNet comparative performance evaluation for image classification

Rinat Mukhometzianov, Juan Antonio Cabrera Carrillo|arXiv (Cornell University)|May 28, 2018

Advanced Neural Network Applications22 references79 citations

TL;DR

The paper evaluates CapsNet against Fisherfaces, LeNet, and ResNet on four datasets, finding CapsNet requires significant compute and underperforms on average, though it may be promising with more robust resources and refined architectures.

ABSTRACT

Image classification has become one of the main tasks in the field of computer vision technologies. In this context, a recent algorithm called CapsNet that implements an approach based on activity vectors and dynamic routing between capsules may overcome some of the limitations of the current state of the art artificial neural networks (ANN) classifiers, such as convolutional neural networks (CNN). In this paper, we evaluated the performance of the CapsNet algorithm in comparison with three well-known classifiers (Fisher-faces, LeNet, and ResNet). We tested the classification accuracy on four datasets with a different number of instances and classes, including images of faces, traffic signs, and everyday objects. The evaluation results show that even for simple architectures, training the CapsNet algorithm requires significant computational resources and its classification performance falls below the average accuracy values of the other three classifiers. However, we argue that CapsNet seems to be a promising new technique for image classification, and further experiments using more robust computation resources and re-fined CapsNet architectures may produce better outcomes.

Motivation & Objective

• Assess CapsNet performance relative to established classifiers on diverse image datasets.
• Investigate computational requirements of training CapsNet versus traditional CNN-based models.
• Provide insights into the potential advantages of CapsNet and directions for future research.

Proposed method

• Compare CapsNet to Fisherfaces, LeNet, and ResNet on four image datasets (faces, traffic signs, objects).
• Measure classification accuracy and training resource requirements.
• Analyze how CapsNet scales with simple architectures and resource constraints.

Experimental results

Research questions

• RQ1 How does CapsNet classification accuracy compare to Fisherfaces, LeNet, and ResNet across multiple datasets?
• RQ2 What are the computational resource requirements for training CapsNet relative to the other classifiers?
• RQ3 Do CapsNet architectures with simple designs achieve competitive performance, and what improvements might be needed for better results?

Key findings

• CapsNet training requires significant computational resources.
• CapsNet's average classification accuracy falls below the other three classifiers on the tested datasets.
• CapsNet appears promising for image classification, with potential improvements from more robust resources and refined architectures.