Eggshapes: A collection of egg-shaped objects and their boundaries

Egg shapes

A collection of egg-shaped objects and their boundaries.

Creators

• Vincent Pelletier
• Jiří Hladůvka

Introduction

Egg shapes are generalized ellipses with positively weighted foci [1]. They can describe a wide range of natural and artificial objects that deviate from the standard ellipse model, such as eggs, avocados, leaves, and rackets.

Fitting egg shapes to these objects is a challenging task that may find many applications in computer vision, image analysis, and pattern recognition. However, there is a lack of publicly available collections of real egg-shaped objects that can be used for developing and testing egg shape-fitting algorithms.

This collection aims to fill this gap by providing a diverse set of images, segmentation masks, and boundaries of real egg-shaped objects.

Context and methodology

The purpose of this collection is to:

• demonstrate the existence of objects that are better described by egg shapes rather than by ellipses, and
• serve as a test collection for the development of egg shape-fitting algorithms.

Diversity and quality

This collection consists of seven datasets:

• eggs_whole: 1,100 whole-egg photographs and segmentation masks from the Egg-segmentation Dataset [2]. We augmented them with 1,100 boundary-coordinates text files.
• eggs_boiled: Images of longitudinally halved, hard-boiled eggs found on the internet were manually segmented, yielding 12 boundaries each for the egg whites and yolks.
• avocados: Images of longitudinally halved avocados were found on the internet. The shells of these avocados were manually segmented, resulting in 6 boundaries. Some of them are slightly deformed.
• leaves: Leaves from trees and plants were deliberately selected and photographed by authors. The selection criteria included being longitudinally symmetrical, egg-shaped, elongated, and possibly pointed. Manual segmentation excluded the stems and produced 23 boundaries.
• cells: Palisade cells of Arabidopsis thaliana in a micro-CT cross-section slice from Water's Gateway to Heaven project were manually segmented, resulting in 159, mostly elliptic boundaries.
• household: 11 spoon heads and 2 toilet seats were photographed and segmented by the authors.
• rackets: Images of tennis, badminton, and squash racket heads sourced from the internet were segmented, resulting in 12 boundaries. The outer shell of the squash head is noted to be pointed. One of the tennis heads is elliptic.

The collection contains a total of 1,129 image+mask pairs and 1,337 boundaries. The images vary in size, resolution, quality, and background. The shapes vary in aspect ratio, eccentricity, curvature, and smoothness.

While the collection covers a range of natural and artificial domains, it is by no means complete. It can be expected that the set of real-world egg shapes is much broader.

The collection may have some biases and limitations, such as the subjective selection of objects and the manual segmentation of masks.

Organization of files

This collection contains files of three types:

1. jpg: photographs,
2. png: segmentation masks, and
3. txt: boundary coordinates.

Each of the seven datasets has photos and segmentation masks stored in:

• images/{dataset}/{photo}.jpg - photo
• images/{dataset}/{photo}.png - segmentation mask

Several objects may be segmented from one image. This means one or several boundaries are associated with one image. These boundaries are stored in:

• boundaries/{dataset}/{photo}_{bID}.txt

where {bID} is a 3-digit boundary identifier.

Acknowledgments

• Water's Gateway to Heaven project for providing the leaf scan,
• Táňa Hladůvková for help with collecting the leaves.

Access

• The collection can be retrieved as a zip file via the following DOI: 10.48436/de66n-2pj41.

License

Creative Commons Attribution-ShareAlike 4.0 International.

References

[1] Gabdulkhakova, A., Kropatsch, W.G.: Generalized conics with the sharp corners. In: Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications. pp. 419–429, (2021), DOI:10.1007/978-3-030-93420-0_39.

[2] Nho, T.: Egg-segmentation Dataset. In: Roboflow Universe, (2023). https://universe.roboflow.com/the-nho/egg-segmentation.

[3] Hladůvka, J. and Kropatsch, W. G. Fitting egg-shapes to discretized object boundaries. In: Discrete geometry and mathematical morphology, pp. 107–119,  (2024), DOI:10.1007/978-3-031-57793-2_9.