pith. sign in

arxiv: 2504.16684 · v1 · submitted 2025-04-23 · 💻 cs.CV · cs.LG

SemanticSugarBeets: A Multi-Task Framework and Dataset for Inspecting Harvest and Storage Characteristics of Sugar Beets

Gerardus Croonen , Andreas Trondl , Julia Simon , Daniel Steininger This is my paper

Pith reviewed 2026-05-22 17:35 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords sugar beetsemantic segmentationobject detectiondatasetcomputer visionpost-harvest inspectionstorage qualitymulti-task learning
0
0 comments X

The pith

Researchers built a dataset and two-stage model to detect sugar beets and segment damage, rot, soil adhesion, and excess vegetation from single RGB images.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a specialized dataset of annotated sugar beet photographs and a computer vision system designed to find the beets and break down their visible problems into categories. These problems include physical damage, rotting areas, soil stuck to the surface, and leftover green parts that can lead to sugar loss while the beets sit in storage before processing. By testing different model sizes and types, the work shows that such automated inspection is feasible and could help processors spot quality issues more quickly and consistently than manual checks alone.

Core claim

The authors claim that their annotated dataset and two-stage framework support accurate detection of sugar beets along with semantic segmentation of damages, rot, soil adhesion, and excess vegetation in monocular RGB images, with experiments confirming strong performance under various conditions.

What carries the argument

The two-stage pipeline for object detection followed by semantic segmentation on a custom high-quality annotated dataset of post-harvest sugar beet images.

If this is right

  • High detection performance enables reliable localization of sugar beets in images for further analysis.
  • Segmentation of fine-grained features provides indicators for potential sugar loss factors like rot and soil.
  • Mass estimation component adds quantitative data to the visual inspection process.
  • Ablation studies reveal how image resolution and environmental factors affect model reliability.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could support integration into automated sorting systems on farms or in factories to reduce waste.
  • Future work might compare visual predictions directly with lab-measured sugar content to strengthen the quality links.
  • Similar multi-task setups may prove useful for inspecting other stored root crops affected by similar issues.

Load-bearing premise

Expert visual annotations on monocular RGB images provide reliable ground truth for fine-grained categories like damages, rot, soil adhesion, and excess vegetation that directly relate to sugar loss.

What would settle it

An experiment that measures actual sugar content chemically in beets the model labels as high versus low in rot or soil adhesion would test whether the visual categories predict real sugar loss.

Figures

Figures reproduced from arXiv: 2504.16684 by Andreas Trondl, Daniel Steininger, Gerardus Croonen, Julia Simon.

Figure 1
Figure 1. Figure 1: Representative images and semantic segmentation masks [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative samples of annotated lighting and soil [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distributions of annotated pixels per processing stage for [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of our two-stage approach to detection and segmentation of sugar beets. First, beets are isolated through instance [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Scatter plot of segmentation performance (vertical axis) [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Fine-grained segmentation performance by meta [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Representative results on test-set images for each processing stage. The leftmost images show object contours provided by [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of normalized bounding-box centers of all [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Representative examples of annotated reference ob [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Annotation-synthesis pipeline to convert original annotations to instance-segmentation annotations of entire sugar beets. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

While sugar beets are stored prior to processing, they lose sugar due to factors such as microorganisms present in adherent soil and excess vegetation. Their automated visual inspection promises to aide in quality assurance and thereby increase efficiency throughout the processing chain of sugar production. In this work, we present a novel high-quality annotated dataset and two-stage method for the detection, semantic segmentation and mass estimation of post-harvest and post-storage sugar beets in monocular RGB images. We conduct extensive ablation experiments for the detection of sugar beets and their fine-grained semantic segmentation regarding damages, rot, soil adhesion and excess vegetation. For these tasks, we evaluate multiple image sizes, model architectures and encoders, as well as the influence of environmental conditions. Our experiments show an mAP50-95 of 98.8 for sugar-beet detection and an mIoU of 64.0 for the best-performing segmentation model.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces the SemanticSugarBeets dataset and a two-stage multi-task framework for detecting sugar beets, performing semantic segmentation of fine-grained attributes (damages, rot, soil adhesion, excess vegetation), and estimating mass from monocular RGB images. It reports extensive ablations over model architectures, image resolutions, encoders, and environmental conditions, with peak results of mAP50-95 = 98.8 for detection and mIoU = 64.0 for segmentation.

Significance. If the annotated visual categories prove reliable proxies for sugar loss, the dataset and framework could support automated quality inspection during harvest and storage, potentially reducing post-harvest losses in sugar production. The systematic ablation study across architectures and conditions, together with the release of a new annotated dataset, constitutes a concrete contribution to agricultural computer vision. The high detection score is a clear strength; the moderate segmentation mIoU indicates that fine-grained tasks remain challenging.

major comments (2)
  1. [Abstract, Introduction, Experiments] Abstract and Introduction: The motivating claim that automated inspection of damages, rot, soil adhesion and excess vegetation will 'aide in quality assurance and thereby increase efficiency' and reduce sugar loss requires that the expert-labeled categories on RGB images are reliable proxies for actual sugar loss or microbial load. No correlation analysis, chemical assay validation, or quantitative mass-estimation results linking segmented regions to measured sugar content are provided, leaving the downstream utility claim unsupported.
  2. [Abstract] Abstract: The abstract states that the method includes mass estimation yet reports no quantitative metrics, error statistics, or ablation results for this task, while supplying concrete numbers only for detection and segmentation. This omission makes the performance claims for the full multi-task framework incomplete.
minor comments (2)
  1. [Abstract, Experiments] The abstract and Experiments section should state the total number of images, number of annotated instances per class, train/validation/test split sizes, and the annotation protocol (number of experts, inter-annotator agreement) so that the reported mAP and mIoU values can be properly contextualized.
  2. [Experiments, Tables/Figures] Table and figure captions should explicitly indicate which model/encoder combination achieves the reported mAP50-95 = 98.8 and mIoU = 64.0, and whether these figures are obtained on the same test split used for all ablations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. We address the major comments point by point below. Revisions have been made to improve clarity and completeness without overstating the current results.

read point-by-point responses

  1. Referee: [Abstract, Introduction, Experiments] Abstract and Introduction: The motivating claim that automated inspection of damages, rot, soil adhesion and excess vegetation will 'aide in quality assurance and thereby increase efficiency' and reduce sugar loss requires that the expert-labeled categories on RGB images are reliable proxies for actual sugar loss or microbial load. No correlation analysis, chemical assay validation, or quantitative mass-estimation results linking segmented regions to measured sugar content are provided, leaving the downstream utility claim unsupported.

    Authors: We agree that the manuscript does not contain direct correlation analysis, chemical assays, or quantitative links between the segmented visual attributes and measured sugar loss or microbial load. The attribute categories were developed in consultation with agricultural domain experts as established visual proxies for known contributors to post-harvest sugar degradation. Our primary contribution is the dataset and multi-task framework for reliable visual detection and segmentation. In the revised manuscript we have adjusted the wording in the abstract and introduction to present these categories as expert-defined visual indicators rather than proven direct predictors of sugar content. We have also added a dedicated limitations paragraph noting that agronomic validation studies correlating the visual outputs with chemical measurements remain future work. revision: yes

  2. Referee: [Abstract] Abstract: The abstract states that the method includes mass estimation yet reports no quantitative metrics, error statistics, or ablation results for this task, while supplying concrete numbers only for detection and segmentation. This omission makes the performance claims for the full multi-task framework incomplete.

    Authors: The referee correctly identifies an inconsistency. Although the two-stage pipeline includes a mass-estimation head that operates on the detection and segmentation outputs, the reported experiments emphasized the novel fine-grained segmentation task and therefore omitted explicit numerical results for mass estimation in the abstract. In the revised version we have updated the abstract to include quantitative performance figures for mass estimation (mean absolute percentage error and associated ablations) and expanded the experiments section with the corresponding results and analysis. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical metrics measured directly against new annotations

full rationale

The paper introduces a novel annotated dataset of sugar beet images and evaluates standard detection (mAP50-95) and segmentation (mIoU) models on it. Reported performance figures are computed as direct comparisons between model outputs and the expert-provided ground-truth labels on the collected data. No equations, parameter fits, or derivations are presented that reduce the claimed results to quantities defined by the same fitted values or self-referential definitions. The evaluation chain is self-contained as standard supervised learning against external annotations, with no load-bearing self-citations or imported uniqueness results.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The claims rest on a newly collected and human-annotated image dataset plus standard supervised deep-learning training assumptions; no new physical constants, particles, or theoretical entities are introduced.

free parameters (1)
  • image resolution and model architecture selection
    Multiple resolutions, backbones, and encoders were tested and the best retained based on validation performance on the collected data.
axioms (1)
  • domain assumption Expert-provided pixel labels accurately reflect the semantic categories of damage, rot, soil, and vegetation in the images.
    All segmentation metrics presuppose that the human annotations constitute reliable ground truth.

pith-pipeline@v0.9.0 · 5696 in / 1193 out tokens · 66252 ms · 2026-05-22T17:35:59.509671+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages · 1 internal anchor

  1. [1]

    A sugar beet leaf disease classification method based on image processing and deep learning

    Kemal Adem, Mehmet Metin Ozguven, and Ziya Altas. A sugar beet leaf disease classification method based on image processing and deep learning. Multimedia Tools and Appli- cations, 82(8):12577–12594, 2023. 2

    work page 2023

  2. [2]

    A vision based row detection system for sugar beet.Comput- ers and Electronics in Agriculture, 60(1):87–95, 2008

    Tijmen Bakker, Hendrik Wouters, Kees Van Asselt, Jan Bontsema, Lie Tang, Joachim M¨uller, and Gerrit van Straten. A vision based row detection system for sugar beet.Comput- ers and Electronics in Agriculture, 60(1):87–95, 2008. 2

    work page 2008

  3. [3]

    A comprehensive cotton leaf disease dataset for enhanced detection and classification

    Prayma Bishshash, Asraful Sharker Nirob, Habibur Shikder, Afjal Hossan Sarower, Touhid Bhuiyan, and Sheak Rashed Haider Noori. A comprehensive cotton leaf disease dataset for enhanced detection and classification. Data in Brief, 57:110913, 2024. 2

    work page 2024

  4. [4]

    Agri- cultural robot dataset for plant classification, localization and mapping on sugar beet fields

    Nived Chebrolu, Philipp Lottes, Alexander Schaefer, Wera Winterhalter, Wolfram Burgard, and Cyrill Stachniss. Agri- cultural robot dataset for plant classification, localization and mapping on sugar beet fields. The International Journal of Robotics Research, 36(10):1045–1052, 2017. 2

    work page 2017

  5. [5]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In Conference on Computer Vision and Pattern Recognition, pages 248–255, 2009. 5

    work page 2009

  6. [6]

    Discriminative unsu- pervised feature learning with exemplar convolutional neu- ral networks

    Alexey Dosovitskiy and Thomas Brox. Discriminative unsu- pervised feature learning with exemplar convolutional neu- ral networks. In IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 1734–1740. IEEE, 2015. 5

    work page 2015

  7. [7]

    Radoslaw M. K. Dziugaite, Daniel M. Roy, et al. Regnet: Designing high-performance neural networks with automat- able structural search. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1822–1830. IEEE, 2020. 5

    work page 2020

  8. [8]

    The pascal visual object classes challenge: A retrospective

    Mark Everingham, SM Ali Eslami, Luc Van Gool, Christo- pher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes challenge: A retrospective. In- ternational Journal of Computer Vision, 111:98–136, 2015. 5

    work page 2015

  9. [9]

    Deep convolutional neu- ral networks for image-based convolvulus sepium detection in sugar beet fields

    Junfeng Gao, Andrew P French, Michael P Pound, Yong He, Tony P Pridmore, and Jan G Pieters. Deep convolutional neu- ral networks for image-based convolvulus sepium detection in sugar beet fields. Plant Methods, 16:1–12, 2020. 2

    work page 2020

  10. [10]

    Automated identification of sugar beet diseases using smartphones

    L Hallau, Marion Neumann, B Klatt, B Kleinhenz, T Klein, C Kuhn, M R¨ohrig, Christian Bauckhage, Kristian Kersting, A-K Mahlein, et al. Automated identification of sugar beet diseases using smartphones. Plant Pathology, 67(2):399– 410, 2018. 2

    work page 2018

  11. [11]

    A cross-domain challenge with panoptic segmentation in agri- culture

    Michael Halstead, Patrick Zimmer, and Chris McCool. A cross-domain challenge with panoptic segmentation in agri- culture. The International Journal of Robotics Research ,

    work page

  12. [12]

    Susceptibility to root tip breakage increases storage losses of sugar beet genotypes

    Christa M Hoffmann and Katharina Schnepel. Susceptibility to root tip breakage increases storage losses of sugar beet genotypes. Sugar Ind, 141(10):625–632, 2016. 1

    work page 2016

  13. [13]

    Search- ing for mobilenetv3

    Andrew Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Wei Wang, Yuan Wu, Piyush Agrawal, Jonathan Bradbury, David Dohan, Jon Shlens, et al. Search- ing for mobilenetv3. In Proceedings of the IEEE Interna- tional Conference on Computer Vision (ICCV), pages 1314–

    work page

  14. [14]

    Segmentation models pytorch

    Pavel Iakubovskii. Segmentation models pytorch. www. github . com / qubvel / segmentation _ models . pytorch, 2019. Accessed: 2024-08-29. 4

    work page 2019

  15. [15]

    Weed detection in sugar beet fields using machine vision

    Abdolabbas Jafari, Seyed Saeid Mohtasebi, Hasan Eghbali Jahromi, and Mahmoud Omid. Weed detection in sugar beet fields using machine vision. International Journal of Agri- culture and Biology, 8(5):602–605, 2006. 2

    work page 2006

  16. [16]

    Spagri-ai: Smart precision agriculture dataset of aerial images at different heights for crop and weed detec- tion using super-resolution

    Martin Jonak, Jan Mucha, Stepan Jezek, Daniel Kovac, and Kornel Cziria. Spagri-ai: Smart precision agriculture dataset of aerial images at different heights for crop and weed detec- tion using super-resolution. Agricultural Systems, 216, 2024. 2

    work page 2024

  17. [17]

    Multiattention network for semantic segmentation of fine-resolution remote sensing images

    Rui Li, Shunyi Zheng, Ce Zhang, Chenxi Duan, Jianlin Su, Libo Wang, and Peter M Atkinson. Multiattention network for semantic segmentation of fine-resolution remote sensing images. IEEE Transactions on Geoscience and Remote Sens- ing, 60:1–13, 2021. 5

    work page 2021

  18. [18]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Proceedings of the European Conference on Computer Vi- sion (ECCV), pages 740–755. Springer, 2014. 5

    work page 2014

  19. [19]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  20. [20]

    An effective classification system for separating sugar beets and weeds for precision farming applications

    Philipp Lottes, Markus Hoeferlin, Slawomir Sander, M M¨uter, P Schulze, and Lammers C Stachniss. An effective classification system for separating sugar beets and weeds for precision farming applications. In International Confer- ence on Robotics and Automation (ICRA), pages 5157–5163. IEEE, 2016. 2

    work page 2016

  21. [21]

    Spec- tral signatures of sugar beet leaves for the detection and dif- ferentiation of diseases

    A-K Mahlein, U Steiner, H-W Dehne, and E-C Oerke. Spec- tral signatures of sugar beet leaves for the detection and dif- ferentiation of diseases. Precision Agriculture, 11:413–431,

    work page

  22. [22]

    Real- time blob-wise sugar beets vs weeds classification for mon- itoring fields using convolutional neural networks

    Andres Milioto, Philipp Lottes, and Cyrill Stachniss. Real- time blob-wise sugar beets vs weeds classification for mon- itoring fields using convolutional neural networks. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 4:41–48, 2017. 2

    work page 2017

  23. [23]

    A patch-image based classification approach for detection of weeds in sugar beet crop

    S Imran Moazzam, Umar S Khan, Waqar S Qureshi, Mohsin I Tiwana, Nasir Rashid, Waleed S Alasmary, Javaid Iqbal, and Amir Hamza. A patch-image based classification approach for detection of weeds in sugar beet crop. IEEE Access, 9, 2021. 2

    work page 2021

  24. [24]

    Using deep learning for image-based plant disease detection

    Sharada P Mohanty, David P Hughes, and Marcel Salath ´e. Using deep learning for image-based plant disease detection. Frontiers in plant science, 7:1419, 2016. 2

    work page 2016

  25. [25]

    Sugar beet damage detection during harvesting using differ- ent convolutional neural network models

    Abozar Nasirahmadi, Ulrike Wilczek, and Oliver Hensel. Sugar beet damage detection during harvesting using differ- ent convolutional neural network models. Agriculture, 11 (11):1111, 2021. 2

    work page 2021

  26. [26]

    Deep learning-based precision agricul- ture through weed recognition in sugar beet fields

    Amin Nasiri, Mahmoud Omid, Amin Taheri-Garavand, and Abdolabbas Jafari. Deep learning-based precision agricul- ture through weed recognition in sugar beet fields. Sustain- able Computing: Informatics and Systems, 35, 2022. 2 9

    work page 2022

  27. [27]

    Automatic de- tection and classification of leaf spot disease in sugar beet using deep learning algorithms

    Mehmet Metin Ozguven and Kemal Adem. Automatic de- tection and classification of leaf spot disease in sugar beet using deep learning algorithms. Physica A: Statistical Me- chanics and Its Applications, 535, 2019. 2

    work page 2019

  28. [28]

    Determination of sucrose content in sugar beet by portable visible and near-infrared spectroscopy

    Leiqing Pan, Qibing Zhu, Renfu Lu, and J Mitchell McGrath. Determination of sucrose content in sugar beet by portable visible and near-infrared spectroscopy. Food chemistry, 167: 264–271, 2015. 2

    work page 2015

  29. [29]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Inter- vention (MICCAI), pages 234–241. Springer, 2015. 5

    work page 2015

  30. [30]

    Beyond map: Towards practical object detection for weed spraying in pre- cision agriculture

    Adrian Salazar-Gomez, Madeleine Darbyshire, Junfeng Gao, Elizabeth I Sklar, and Simon Parsons. Beyond map: Towards practical object detection for weed spraying in pre- cision agriculture. In 2022 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS), pages 9232–

    work page 2022

  31. [31]

    Scalabel open-source web annotation tool

    Scalabel. Scalabel open-source web annotation tool. www. github.com/scalabel/scalabel, 2024. Accessed: 2024-08-19. 3

    work page 2024

  32. [32]

    Remote de- tection of rhizomania in sugar beets

    K Steddom, G Heidel, D Jones, and CM Rush. Remote de- tection of rhizomania in sugar beets. Phytopathology, 93(6): 720–726, 2003. 2

    work page 2003

  33. [33]

    The cropandweed dataset: A multi-modal learning approach for efficient crop and weed manipulation

    Daniel Steininger, Andreas Trondl, Gerardus Croonen, Julia Simon, and Verena Widhalm. The cropandweed dataset: A multi-modal learning approach for efficient crop and weed manipulation. In Proceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision , pages 3729– 3738, 2023. 2

    work page 2023

  34. [34]

    Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations

    Carole H Sudre, Wenqi Li, Tom Vercauteren, Sebastien Ourselin, and M Jorge Cardoso. Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support , pages 240–248. Springer, 2017. 5

    work page 2017

  35. [35]

    Mingxing Tan and Quoc V . Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In Pro- ceedings of the 36th International Conference on Machine Learning (ICML), pages 6105–6114. PMLR, 2019. 5

    work page 2019

  36. [36]

    Phenobench: A large dataset and benchmarks for semantic image interpreta- tion in the agricultural domain

    Jan Weyler, Federico Magistri, Elias Marks, Yue Linn Chong, Matteo Sodano, Gianmarco Roggiolani, Nived Che- brolu, Cyrill Stachniss, and Jens Behley. Phenobench: A large dataset and benchmarks for semantic image interpreta- tion in the agricultural domain. IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2024. 2

    work page 2024

  37. [37]

    Deep learning for non-invasive diagnosis of nutrient deficiencies in sugar beet using rgb images

    Jinhui Yi, Lukas Krusenbaum, Paula Unger, Hubert H ¨uging, Sabine J Seidel, Gabriel Schaaf, and Juergen Gall. Deep learning for non-invasive diagnosis of nutrient deficiencies in sugar beet using rgb images. Sensors, 20(20):5893, 2020. 2

    work page 2020

  38. [38]

    Mobileone: Ultra-lightweight and efficient neural networks for mobile devices

    Peng Zhang, Wei Liu, Yunhe Zhang, Shuang Zhang, et al. Mobileone: Ultra-lightweight and efficient neural networks for mobile devices. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) , pages 16231–16240. IEEE, 2021. 5

    work page 2021

  39. [39]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2881–2890, 2017. 5 10 SemanticSugarBeets: A Multi-Task Framework and Dataset for Inspecting Harvest and Storage Characteristics of Sugar Beets (Supplementary Mater...

    work page 2017

  40. [40]

    Compose a group of beets (3 to 5) to fit inside the camera frame, held in landscape mode

    work page

  41. [41]

    Put a folding ruler (or other object of known size) in the frame, ensuring its full visibility

    work page

  42. [42]

    From a standing position, take two (almost identical) photographs from a top-view perspective

    work page

  43. [43]

    Flip the beets and put them back in roughly the same position

    work page

  44. [44]

    This photo will also allow for the quick identification of separate beet groups and beet sides when viewing and meta-annotating the photos

    Force a camera refocus by taking a photograph of a nearby object, such as your hand. This photo will also allow for the quick identification of separate beet groups and beet sides when viewing and meta-annotating the photos

    work page

  45. [45]

    Repeat steps 2-3. B. Extended dataset analysis Tab. 5 provides a complete list of recording sessions and corresponding statistics and meta-parameters. The distri- bution of bounding box centers across all beet instances is depicted in Fig. 9. Representative examples for both classes of annotated reference markers are visualized in Fig. 10. C. Extended met...

    work page