One-Shot Identification with Different Neural Network Approaches

TL;DR

The paper addresses one-shot identification under severe data scarcity by evaluating three architectures: a classic CNN with merged images, a Siamese network, and a Siamese network incorporating Capsule networks (CapsNet). It demonstrates that stacking input channels often outperforms simple image merging, while CapsNet-based Siamese networks achieve the best accuracy across industrial and face recognition tasks; data augmentation and decoder-enabled synthesis further boost performance. The approach is validated on an industrial dataset, smallNORB, and the AT&T Faces dataset, showing CapsNet-based methods are especially effective in very small data regimes and offer potential for explainability and continual learning. These results suggest practical, data-efficient strategies for one-shot and zero-shot identification in real-world scenarios.

Abstract

Convolutional neural networks (CNNs) have been widely used in the computer vision community, significantly improving the state-of-the-art. But learning good features often is computationally expensive in machine learning settings and is especially difficult when there is a lack of data. One-shot learning is one such area where only limited data is available. In one-shot learning, predictions have to be made after seeing only one example from one class, which requires special techniques. In this paper we explore different approaches to one-shot identification tasks in different domains including an industrial application and face recognition. We use a special technique with stacked images and use siamese capsule networks. It is encouraging to see that the approach using capsule architecture achieves strong results and exceeds other techniques on a wide range of datasets from industrial application to face recognition benchmarks while being easy to use and optimise.

Figures (12)

• Figure 1: General approach to one-shot identification using merged images. Two images are merged and then put into a CNN to classify whether they contain the same or different objects. The images on the left are a random selection from the AT&T dataset used in Section \ref{['sec:Experiments']}.
• Figure 2: Merging two images into a bigger image by joining them horizontally or vertically.Mohr21
• Figure 3: Merging two images converted to greyscale by stacking them resulting in a two-channel image.Mohr21
• Figure 4: Architecture of the Siamese Network which is used as a baselineDeshpande.2020
• Figure 5: Architecture of the used CapsNet.Quetscher.2022