Online Estimation of Table-Top Grown Strawberry Mass in Field Conditions with Occlusions

Hui Zhao; Jinshan Zhen; Tianxiao Zhu; Ya Xiong; Yuanyue Ge

REVIEW 3 major objections 6 minor 22 references

Online Estimation of Table-Top Grown Strawberry Mass in Field Conditions with Occlusions

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A camera-only pipeline estimates strawberry mass with mean errors of 8.11% for isolated fruit and 10.47% for occluded fruit, using segmentation, CycleGAN completion, and tilt correction.

desk verdict A sensible engineering pipeline for strawberry mass estimation, but the headline 8.11%/10.47% errors may be in-sample fit residuals, not out-of-sample predictions. read the letter →

arxiv 2507.23487 v1 pith:EBV27C3Z submitted 2025-07-31 cs.CV cs.RO

Jinshan Zhen , Yuanyue Ge , Tianxiao Zhu , Hui Zhao , Ya Xiong This is my paper · ORCID

classification cs.CVcs.RO

keywords strawberrymassestimationRGB-DcameraCycleGANocclusioncompletionYOLOv8-Seginstancesegmentationtilt-anglecorrectionpolynomialregressionnon-destructivefruitweighing

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that strawberry mass can be estimated online, without touching the fruit, from a single RGB-D camera view even when berries are partly hidden by leaves, stems, or neighboring fruit. The proposed pipeline segments each berry, regenerates the hidden portion with a CycleGAN, corrects the frontal projection for the berry's tilt, and maps projected area to volume through a cubic polynomial. Against field-collected samples the mean mass error is 8.11% for isolated strawberries and 10.47% for occluded ones. If those figures hold, mass-based grading and robotic harvesting could be driven by vision alone.

What carries the argument

Three components carry the argument: (1) CycleGAN, an unpaired image-to-image translation network whose cycle-consistency loss lets it map occluded strawberry images to complete ones without paired training data; (2) a pose estimator that locates the stem, tip, and belly convexity of the berry, fits a plane to local convex points, derives the tilt angle $\theta$ of the central axis, and multiplies the visible area by $\cos\theta$ to recover the frontal projection; and (3) a cubic polynomial regression from maximum cross-sectional area to volume, fitted on field samples with $R^2 = 0.9037$, combined with the average density of ripe strawberries to yield mass. The polynomial absorbs the nonlinear area-to-volume relation that makes smaller berries particularly sensitive to area error.

What would settle it

Scan occluded strawberries with a method that captures their true hidden geometry, such as multi-view reconstruction or weighing immediately after picking, and compare against the volume produced by the CycleGAN-plus-average-depth pipeline; if mass errors correlate with occlusion fraction or with the depth difference between occluded and isolated berries, the fill heuristic is the culprit.

Watch

Extended reading notes

Core claim

The central claim is that occlusion need not prevent accurate mass estimation: YOLOv8-Seg isolates each fruit, a CycleGAN completes the occluded region with cycle-consistency training between occluded and intact image domains, and a symmetry-based pose correction converts the visible projection into a true frontal area. A cubic polynomial $y = -24.9926 + 7.1919x - 0.3063x^2 + 0.0052x^3$ then predicts volume from that area, with mass following from average strawberry density. The paper reports mean mass errors of 8.11% for isolated and 10.47% for occluded strawberries, and an occlusion-recovery stage that beats the LaMa inpainting model in pixel-area ratio (0.978 vs 1.112) and in the fraction of reconstructions with IoU in [0.9, 1] (92.3% vs 47.7%).

Load-bearing premise

The entire error budget rests on the assumption that the depth hidden behind an occluding leaf, stem, or berry can be replaced by the average depth of isolated strawberries, an approximation the paper states but never validates or bounds.

Editorial extensions

If this is right

One RGB-D frame suffices for mass-based grading: the paper's grade thresholds (A > 30 g, B 20-30 g, C 10-20 g, D ≤10 g) become computable in real time at 110-200 ms per frame.
Occlusion no longer forces either geometric shape guessing or destructive weighing, because the CycleGAN stage handles hidden regions that kite-geometry and symmetry-plane predecessors cannot.
The tilt correction removes a systematic source of underestimation for naturally slanted berries, which is consistent with the isolated-fruit area error staying at 4.14%.
The error structure across stages (area 4-6%, volume 7-10%, mass 8-10.5%) identifies where each part of the pipeline loses accuracy, giving a clear target for further improvement.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The average-depth fill for occluded regions is the most fragile step: the paper never verifies that an occluded berry's hidden depth equals the mean depth of isolated berries, so systematic distance differences would bias volume and hence mass.
Because mass is volume times a fixed average density, variety, ripeness, and moisture content will shift the error; a per-class or per-ripeness density calibration is a natural extension.
The segment-complete-correct-regress skeleton should transfer to other irregular, mutually occluding fruits such as raspberries or cherry tomatoes, where the same depth-fill question would arise.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: the pipeline is sensible and the components are well chosen, but the two headline numbers rest on a polynomial fit that appears to be evaluated on the same field samples it was fitted to. No held-out split is described for the regression in Section III-C, so the 8.11% and 10.47% mean errors are likely training residuals, not out-of-sample predictions.

What is actually new: applying CycleGAN to complete occluded strawberry regions and then using the completed mask with a tilt-angle correction for volume/mass. The CycleGAN vs LaMa comparison on 325 test images with PAR and IoU is a real evaluation, even if those metrics are not linked to the final mass error. The segmentation result (mAP 0.91) is solid. Credit where due: the Discussion admits the polynomial regression is sensitive to data distribution and that a fixed density ignores variety, maturity, and moisture. That is the right kind of honest limitation statement.

The soft spots, in proportion. The load-bearing problem is the in-sample fit. Field samples are collected, Eq. 6 is fitted, and Tables II–III report the resulting errors. Without a separate test split or cross-validation, the mass errors are not trustworthy. This affects both the isolated and occluded numbers because both go through the same cubic. The depth-fill heuristic (average depth of isolated strawberries) is crude and unvalidated; the paper does not bound the error it introduces. There is also no no-restoration baseline, so we cannot see what CycleGAN actually buys for mass estimation. And the IoU/PAR comparison, while a genuine ablation of sorts, is disconnected from the mass metric; a PAR of 0.978 and IoU of 92.3% are good, but we don't know how those translate into the 10.47% error.

Minor: "online" is not demonstrated end-to-end; the per-frame times (110–200 ms) are plausible but there is no full-pipeline latency. The 256x256 resolution cap on CycleGAN is acknowledged as a detail-fidelity limit.

Verdict: a useful engineering paper for people building strawberry harvesters or yield monitors. It deserves peer review, but only after the authors do a proper held-out or cross-validated evaluation of Eq. 6, add a no-restoration baseline, and report error bars or at least sample counts. If those numbers hold on independent data, this is a solid applied contribution. I would send it to review with the expectation of major revision rather than desk-reject.

Referee Report

3 major / 6 minor

Summary. This paper presents a pipeline for non-destructive, online mass estimation of table-top grown strawberries in field conditions, targeting both isolated and occluded fruits. The pipeline consists of YOLOv8-Seg instance segmentation, CycleGAN-based occlusion completion, a geometric tilt-angle correction of the projected frontal area, cubic polynomial regression from cross-sectional area to volume, and conversion to mass using an average strawberry density. Experiments reported mean mass estimation errors of 8.11% for isolated and 10.47% for occluded strawberries, with CycleGAN outperforming LaMa on pixel area ratio and IoU metrics for occlusion recovery. The paper is evaluated on real field data and includes visual comparisons, but the central accuracy claims rest on an area-volume regression that appears to be fitted and evaluated on the same samples.

Significance. If the reported accuracies are reproducible on independent data, the proposed system would be a valuable practical contribution to automated harvesting and yield monitoring in strawberries, particularly because it addresses occlusions with CycleGAN and accounts for pose variations. The use of genuine field-captured RGB-D data and the direct comparison with a state-of-the-art inpainting method (LaMa) are strengths, and the paper includes useful visual verification. However, the current evidence does not yet support the headline error rates: the polynomial regression in Eq. (6) is fit on field samples, and the errors in Tables II–III appear to be computed on the same samples, making them in-sample residuals rather than out-of-sample predictions. The depth-fill heuristic and constant-density assumption are also not quantitatively validated. No code or data are provided, further limiting independent verification.

major comments (3)

[III-C, Eq. (6), Tables II and III] The cubic polynomial regression in Eq. (6) is fit to field samples, and the average errors in Tables II and III are computed on the same experimental data. The 6:3:1 split described in Section II-A applies to the 1,200 images used for YOLOv8-Seg and CycleGAN training, not to the regression samples. There is no statement that the field samples were split into fitting and evaluation subsets, nor any cross-validation. Consequently, the reported 8.11% and 10.47% mean mass errors are likely the residuals of the fit rather than predictions on new strawberries, and the headline accuracy claim in the abstract is not yet supported. Please report the number of field samples, perform a train/test split or cross-validation for the polynomial, and report the out-of-sample errors.
[II-C] In occluded scenarios, the missing depth behind the restored region is filled using 'the average depth of isolated strawberries.' This is a constant-fill heuristic, and no validation or comparison against a no-fill baseline, a per-pixel depth model, or an alternative interpolation is provided. Since this fill is applied to all occluded cases, the reported 10.47% occluded mass error is confounded with the error of this assumption. An ablation that removes or varies the depth-fill is necessary to establish that the occlusion-completion pipeline itself brings the claimed benefit.
[III-C] The mass is calculated by multiplying the estimated volume by 'the average density of ripe strawberries,' but the paper never reports the actual density value, how it was measured, or its variability. Section IV acknowledges that density changes with variety, maturity, and moisture content, so a constant density is a potential source of systematic error. Please report the measured density (with units), the number of samples used to obtain it, and a sensitivity analysis showing how the reported mass errors would change within the observed density range.

minor comments (6)

[III-C] The text states that the measured data points and fitted curves are shown in Fig. 10, but the regression plot is Fig. 9; Figure 10 shows field test images. Please correct the cross-reference.
[IV] The last paragraph of the Discussion contains the typo 'masss estimation'; it should be 'mass estimation.'
[Abstract, III-B, Fig. 6] The abstract and the main text state that in the IoU interval [0.9–1], LaMa achieves 47.7%, while the caption of Fig. 6 says LaMa reaches only 39.7%. These numbers are inconsistent and should be reconciled.
[Table I] The notation in Table I, such as '1 + 0.112' and '1 - 0.13,' is unconventional and ambiguous. Please report the mean and standard deviation (or variance) explicitly, and define how the deviation from the ideal value is computed.
[II-A] The paper reports a validation mAP@0.5 of 0.91 for YOLOv8-Seg but does not state whether the validation set used for this metric is part of the 6:3:1 split described for the 1,200 images, and the numbers of images in the training, testing, and validation subsets are not given.
[III-C] The paper does not state the number of field samples used to fit Eq. (6) and to compute the errors in Tables II and III. Reporting the sample size is necessary to assess the statistical reliability of the error rates.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline mass errors are in-sample residuals: Eq. (6) is fitted on the same field samples whose volume and mass errors are later reported as 'predicted' values.

fitted input called prediction [Section III-C, Eq. (6) and Tables II-III]
"To assess the accuracy of the algorithm, a series of comparative experiments were conducted. Field samples were collected to record the maximum vertical cross-sectional area, tilt angle, volume, and mass of the strawberries. ... In addition, we fitted a cubic polynomial regression model to describe the relationship between the maximum cross-sectional area and the volume of the strawberries: y = −24.9926 + 7.1919 · x − 0.3063 · x2 + 0.0052 · x3 (6)"

The same field samples used to determine the coefficients of Eq. (6) are the ones for which Tables II and III report volume errors of 7.52%/10.07% and mass errors of 8.11%/10.47%. The paper never states that a different subset of strawberries was held out for this regression: the 6:3:1 split in Section II-A is described for the 1200 RGB-D images that train YOLOv8-Seg and CycleGAN, not for the area-to-volume polynomial. Thus, as described, the reported volume and mass errors are in-sample residuals of the fitted cubic (R^2 = 0.9037) evaluated on its own training fruits, so the headline 'mean mass estimation errors of 8.11%...10.47%' is a fitted input presented as a prediction rather than an out-of-sample validation.

full rationale

The central claim in the abstract, a mean mass error of 8.11% for isolated and 10.47% for occluded strawberries, is not self-contained as an out-of-sample result. Section III-C fits a cubic polynomial (Eq. 6) to manually measured field samples and then uses that same equation to compute the volume and mass errors shown in Tables II and III, with no described train/test split or cross-validation for this regression. The 6:3:1 split mentioned earlier applies to the image dataset used for YOLOv8-Seg and CycleGAN training, not to the polynomial regression, so the reported errors are in-sample fit residuals. The CycleGAN versus LaMa comparison is evaluated on 325 test images and does not reduce to its training inputs, so that part is not circular. The depth-fill heuristic for occluded regions and the use of average strawberry density are assumptions that may introduce bias but are not formally circular. The self-citations in the paper (e.g., [3], [13]) describe prior robot and shape-completion work but are not load-bearing for the mass-estimation error claims. On balance, the main prediction claim reduces by construction to a fitted regression, giving a partial circularity score of 6.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central mass claim rests on a fitted polynomial, a batch-average density, and an ad hoc depth-fill constant. None of these are derived from first principles or validated on external data. The regression coefficients are stated in Eq. (6); the density and depth-fill values are not reported numerically. There are no invented physical entities, but the fitting and depth-fill choices are the main burden the reader is asked to accept.

free parameters (5)

Cubic regression coefficients (beta0 to beta3) = -24.9926, 7.1919, -0.3063, 0.0052
Eq. (6) fits the area-to-volume relationship to manually measured samples. These coefficients are the core of the volume estimator.
Average strawberry density = not reported
Mass is obtained as volume times an average density taken from the same strawberry population; density is not measured per fruit.
Depth-fill constant (average depth of isolated strawberries) = not reported
Depth behind occluded regions is replaced by a single average depth value from isolated strawberries, which is a hand-chosen substitution.
Tilt-angle algorithm search radius and neighborhood sizes = r, 30 nearest points, 100 sampled points
These hand-tuned parameters control the convex-hull based tilt estimation; no sensitivity analysis is given.
CycleGAN training hyperparameters = batch size 2, kernel count 96, learning rate 0.0002 to 0.0001, lambda not reported
Training configurations chosen by hand; the cycle-consistency weight lambda is not reported, so exact training is not specified.

assumptions (6)

standard math Least-squares polynomial regression maps frontal projection area to volume.
Eq. (5) and Eq. (6) assume a polynomial relationship between area and volume, which is a standard regression formulation.
standard math Pinhole camera projection with cosine tilt correction recovers the true frontal projection area.
Section II-C uses the symmetry of strawberries and cos(theta) to correct the projection error caused by tilt.
domain assumption Strawberry volume can be predicted from a single corrected frontal projection area.
The entire volume estimation chain (Section II-C) rests on this geometric assumption, which is not validated against multi-view measurements.
domain assumption Depth of the restored occluded region can be approximated by the average depth of isolated strawberries.
Section II-C, last paragraph, states this substitution explicitly without validation or error bounds.
domain assumption Strawberry density is constant across ripeness, variety, and moisture.
The Discussion section notes this assumption and acknowledges that density changes would affect accuracy.
ad hoc to paper CycleGAN-generated images preserve the pixel area and shape of the true fruit closely enough for area-based volume estimation.
Section III-B reports PAR and IoU for the restoration, but does not measure how restoration error propagates into mass estimation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Online Estimation of Table-Top Grown Strawberry Mass in Field Conditions with Occlusions." pith.science (2026). https://pith.science/paper/EBV27C3Z

@misc{pith2026250723487,
  author       = {Pith},
  title        = {Pith review of: Online Estimation of Table-Top Grown Strawberry Mass in Field Conditions with Occlusions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EBV27C3Z}},
  note         = {Machine review of arXiv:2507.23487}
}

read the original abstract

Accurate mass estimation of table-top grown strawberries under field conditions remains challenging due to frequent occlusions and pose variations. This study proposes a vision-based pipeline integrating RGB-D sensing and deep learning to enable non-destructive, real-time and online mass estimation. The method employed YOLOv8-Seg for instance segmentation, Cycle-consistent generative adversarial network (CycleGAN) for occluded region completion, and tilt-angle correction to refine frontal projection area calculations. A polynomial regression model then mapped the geometric features to mass. Experiments demonstrated mean mass estimation errors of 8.11% for isolated strawberries and 10.47% for occluded cases. CycleGAN outperformed large mask inpainting (LaMa) model in occlusion recovery, achieving superior pixel area ratios (PAR) (mean: 0.978 vs. 1.112) and higher intersection over union (IoU) scores (92.3% vs. 47.7% in the [0.9-1] range). This approach addresses critical limitations of traditional methods, offering a robust solution for automated harvesting and yield monitoring with complex occlusion patterns.

Figures

Figures reproduced from arXiv: 2507.23487 by the authors.

**Figure 1.** Pipeline of mass estimation for occluded strawberries: (a) The RGB image is first processed by an instance segmentation network to isolate the strawberry region, followed by occlusion restoration using CycleGAN; (b) The restored strawberry image is projected into a 3D space using a depth map and camera geometry for volume estimation, and the mass is subsequently predicted based on a volume-to-mass regression model d… view at source ↗

**Figure 2.** CycleGAN-based shape completion framework: The generator GAB transforms an occluded strawberry image from domain A into a completed (fake) image in domain B, which is then passed to the discriminator DB for adversarial training against real images in domain B. Meanwhile, the reverse generator GBA reconstructs the original occluded image, and an L2 loss is applied between the input and reconstructed image to enforce… view at source ↗

**Figure 3.** Pose and projection of strawberries: (a) Comparison of the projected visible areas between a vertically aligned strawberry and a tilted strawberry, highlighting how pose affects the observed contour. (b) Illustration of the tilt angle θ formed between the strawberry’s central axis and the vertical viewing direction, which is used for geometric correction in volume estimation. as the central point, representing the l… view at source ↗

Figures from the paper (7 more)

**Figure 5.** Figure 5: Scatter plots of pixel area ratios for CycleGAN and LaMa: Each dot represents the ratio between the pixel area of the restored fruit and the ground truth across 325 test images. The red dashed line denotes the ideal ratio (1.00), while green dashed lines mark acceptabl…

**Figure 4.** Figure 4: Precision–Recall (PR) curves for segmentation under occlusion: The PR curves illustrate the segmentation performance of YOLOv8-Seg on covered (orange), uncovered (light blue), and all samples (blue bold). The model achieved a mean average precision (mAP@0.5) of 0.910 a…

**Figure 6.** Figure 6: IoU distribution comparison between CycleGAN and LaMa: Pie charts visualize the distribution of IoU values for recovered strawberry contours. The IoU range is divided into four intervals: [0.0, 0.6), [0.6, 0.8), [0.8, 0.9), and [0.9, 1.0]. A higher proportion in the [0…

**Figure 8.** Figure 8: , CycleGAN also performed well in texture restoration, effectively filling occluded regions while maintaining global image consistency. In contrast, the regions restored by LaMa often exhibited inconsistencies with the surrounding textures. However, CycleGAN was const…

**Figure 7.** Figure 7: Comparison of shape recovery performance: Visual comparison between CycleGAN and LaMa models in recovering strawberry contours under occlusion. The leftmost column (a-1 to c-1) displays the original RGB images captured in natural field environments. The second column (…

**Figure 9.** Figure 9: Polynomial regression of surface area and volume: A third-order polynomial regression was used to fit the non-linear relationship between the maximum cross-sectional area and the measured volume of strawberries. Blue dots represent actual measurements, and the red curv…

**Figure 10.** Figure 10: Test results under field conditions: Original RGB images of strawberries under varying occlusion and lighting conditions (top row), final system output showing estimated mass and grade in green text (bottom row). of CycleGAN for agricultural occlusion recovery and a s…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 21 canonical work pages

[1]

Fruit sizing in orchard: A review from caliper to machine vision with deep learning

Chiranjivi Neupane, Maisa Pereira, Anand Koirala, and Kerry B Walsh. Fruit sizing in orchard: A review from caliper to machine vision with deep learning. Sensors, 23(8):3868, 2023

work page 2023
[2]

Automation on fruit and vegetable grading system and food traceability

Naoshi Kondo. Automation on fruit and vegetable grading system and food traceability. Trends in F ood Science & Technology , 21(3):145– 152, 2010

work page 2010
[3]

An au- tonomous strawberry-harvesting robot: Design, development, integra- tion, and field evaluation

Ya Xiong, Yuanyue Ge, Lars Grimstad, and P ˚al J From. An au- tonomous strawberry-harvesting robot: Design, development, integra- tion, and field evaluation. Journal of Field Robotics , 37(2):202–224, 2020

work page 2020
[4]

Partial least squares estimation of crop moisture and density by near- infrared spectroscopy

Davide Cassanelli, Nicola Lenzini, Luca Ferrari, and Luigi Rovati. Partial least squares estimation of crop moisture and density by near- infrared spectroscopy. IEEE Transactions on Instrumentation and Measurement, 70:1–10, 2021

work page 2021
[5]

Individual tree diameter estimation in small-scale forest inventory using uav laser scanning

Yuanshuo Hao, Faris Rafi Almay Widagdo, Xin Liu, Ying Quan, Lihu Dong, and Fengri Li. Individual tree diameter estimation in small-scale forest inventory using uav laser scanning. Remote Sensing , 13(1):24, 2020

work page 2020
[6]

A novel hybrid-view technique for accurate mass estimation of kimchi cabbage using computer vision

Hae-Il Yang, Sung-Gi Min, Ji-Hee Yang, Jong-Bang Eun, and Young- Bae Chung. A novel hybrid-view technique for accurate mass estimation of kimchi cabbage using computer vision. Journal of F ood Engineering, 378:112126, 2024

work page 2024
[7]

Assessing automatic data processing algorithms for rgb-d cameras to predict fruit size and weight in apples

Juan C Miranda, Jaume Arn ´o, Jordi Gen ´e-Mola, Jaume Lordan, Luis As´ın, and Eduard Gregorio. Assessing automatic data processing algorithms for rgb-d cameras to predict fruit size and weight in apples. Computers and Electronics in Agriculture , 214:108302, 2023

work page 2023
[8]

A deep learning system for single and overall weight estimation of melons using unmanned aerial vehicle images

Aharon Kalantar, Yael Edan, Amit Gur, and Iftach Klapp. A deep learning system for single and overall weight estimation of melons using unmanned aerial vehicle images. Computers and Electronics in Agriculture, 178:105748, 2020

work page 2020

Show all 22 references

[9]

Strawberry picking point localization ripeness and weight estimation

Alessandra Tafuro, Adeayo Adewumi, Soran Parsa, Ghalamzan E Amir, and Bappaditya Debnath. Strawberry picking point localization ripeness and weight estimation. In 2022 International conference on robotics and automation (ICRA) , pages 2295–2302. Ieee, 2022

2022
[10]

Strawberry weight estimation based on plane-constrained binary division point cloud completion

Yanjiang Huang, Jiepeng Liu, and Xianmin Zhang. Strawberry weight estimation based on plane-constrained binary division point cloud completion. In 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages 11846–11852. IEEE, 2024

2024
[11]

Non- destructive estimation of fruit weight of strawberry using machine learning models

Jayanta Kumar Basak, Bhola Paudel, Na Eun Kim, Nibas Chandra Deb, Bolappa Gamage Kaushalya Madhavi, and Hyeon Tae Kim. Non- destructive estimation of fruit weight of strawberry using machine learning models. Agronomy, 12(10):2487, 2022

2022
[12]

A simple and efficient method for automatic strawberry shape and size estimation and classification

Lin Mar Oo and Nay Zar Aung. A simple and efficient method for automatic strawberry shape and size estimation and classification. Biosystems engineering , 170:96–107, 2018

2018
[13]

Yuanyue Ge, Ya Xiong, and P ˚al J. From. Symmetry-based 3d shape completion for fruit localisation for harvesting robots. Biosystems Engineering, 197:188–202, 2020

2020
[14]

High-fidelity image inpainting with gan inversion

Yongsheng Yu, Libo Zhang, Heng Fan, and Tiejian Luo. High-fidelity image inpainting with gan inversion. In European Conference on Computer Vision, pages 242–258. Springer, 2022

2022
[15]

Hi-gan: Hierarchical inpainting gan with auxiliary inputs for combined rgb and depth inpainting

Ankan Dash, Jingyi Gu, and Guiling Wang. Hi-gan: Hierarchical inpainting gan with auxiliary inputs for combined rgb and depth inpainting. arXiv preprint arXiv:2402.10334 , 2024

2024 arXiv
[16]

Mi-gan: A simple baseline for image inpainting on mobile devices

Andranik Sargsyan, Shant Navasardyan, Xingqian Xu, and Humphrey Shi. Mi-gan: A simple baseline for image inpainting on mobile devices. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 7335–7345, 2023

2023
[17]

Shapemorph: 3d shape completion via blockwise discrete diffusion

Jiahui Li, Pourya Shamsolmoali, Yue Lu, and Masoumeh Zareapoor. Shapemorph: 3d shape completion via blockwise discrete diffusion. pages 2818–2827, 2025

2025
[18]

Diffcomplete: Diffusion-based generative 3d shape completion

Ruihang Chu, Enze Xie, Shentong Mo, Zhenguo Li, Matthias Nießner, Chi-Wing Fu, and Jiaya Jia. Diffcomplete: Diffusion-based generative 3d shape completion. 2023

2023
[19]

Un- paired image-to-image translation using cycle-consistent adversarial networks

Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Un- paired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision , pages 2223–2232, 2017

2017
[20]

Mask r-cnn

Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Girshick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision , pages 2961–2969, 2017

2017
[21]

U-net: Con- volutional networks for biomedical image segmentation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Con- volutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, par...

2015
[22]

Cspnet: A new backbone that can enhance learning capability of cnn

Chien-Yao Wang, Hong-Yuan Mark Liao, Yueh-Hua Wu, Ping-Yang Chen, Jun-Wei Hsieh, and I-Hau Yeh. Cspnet: A new backbone that can enhance learning capability of cnn. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 390–391, 2020

2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Fruit sizing in orchard: A review from caliper to machine vision with deep learning

Chiranjivi Neupane, Maisa Pereira, Anand Koirala, and Kerry B Walsh. Fruit sizing in orchard: A review from caliper to machine vision with deep learning. Sensors, 23(8):3868, 2023

work page 2023

[2] [2]

Automation on fruit and vegetable grading system and food traceability

Naoshi Kondo. Automation on fruit and vegetable grading system and food traceability. Trends in F ood Science & Technology , 21(3):145– 152, 2010

work page 2010

[3] [3]

An au- tonomous strawberry-harvesting robot: Design, development, integra- tion, and field evaluation

Ya Xiong, Yuanyue Ge, Lars Grimstad, and P ˚al J From. An au- tonomous strawberry-harvesting robot: Design, development, integra- tion, and field evaluation. Journal of Field Robotics , 37(2):202–224, 2020

work page 2020

[4] [4]

Partial least squares estimation of crop moisture and density by near- infrared spectroscopy

Davide Cassanelli, Nicola Lenzini, Luca Ferrari, and Luigi Rovati. Partial least squares estimation of crop moisture and density by near- infrared spectroscopy. IEEE Transactions on Instrumentation and Measurement, 70:1–10, 2021

work page 2021

[5] [5]

Individual tree diameter estimation in small-scale forest inventory using uav laser scanning

Yuanshuo Hao, Faris Rafi Almay Widagdo, Xin Liu, Ying Quan, Lihu Dong, and Fengri Li. Individual tree diameter estimation in small-scale forest inventory using uav laser scanning. Remote Sensing , 13(1):24, 2020

work page 2020

[6] [6]

A novel hybrid-view technique for accurate mass estimation of kimchi cabbage using computer vision

Hae-Il Yang, Sung-Gi Min, Ji-Hee Yang, Jong-Bang Eun, and Young- Bae Chung. A novel hybrid-view technique for accurate mass estimation of kimchi cabbage using computer vision. Journal of F ood Engineering, 378:112126, 2024

work page 2024

[7] [7]

Assessing automatic data processing algorithms for rgb-d cameras to predict fruit size and weight in apples

Juan C Miranda, Jaume Arn ´o, Jordi Gen ´e-Mola, Jaume Lordan, Luis As´ın, and Eduard Gregorio. Assessing automatic data processing algorithms for rgb-d cameras to predict fruit size and weight in apples. Computers and Electronics in Agriculture , 214:108302, 2023

work page 2023

[8] [8]

A deep learning system for single and overall weight estimation of melons using unmanned aerial vehicle images

Aharon Kalantar, Yael Edan, Amit Gur, and Iftach Klapp. A deep learning system for single and overall weight estimation of melons using unmanned aerial vehicle images. Computers and Electronics in Agriculture, 178:105748, 2020

work page 2020