Pith. sign in

REVIEW 5 major objections 6 minor 33 references

Weakly Supervised Panoptic Segmentation for Defect-Based Grading of Fresh Produce

T0 review · 5 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Training on SAM-generated masks matches hand-annotated training for banana defect grading.

desk verdict The central SAM-vs-human training comparison is honest and holds up, but the grading-utility numbers (count accuracy, size correlation) rest on a single annotator and need an inter-annotator baseline before they mean what the paper suggests. read the letter →

arxiv 2411.16219 v2 pith:FPF2R6MH submitted 2024-11-25 cs.CV

Manuel Knott , Divinefavour Odion , Sameer Sontakke , Anup Karwa , Thijs Defraeye This is my paper · ORCID
classification cs.CV
keywords weaklysupervisedsegmentationpanopticSegmentAnythingModeldefectgradingbananasurfacedefectslow-dataregimefoundationmodelsagriculturalcomputervision
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 a panoptic segmentation model can be trained on dense masks auto-generated by the Segment Anything Model (SAM) from coarse inputs—bounding boxes around defects and points on bananas—instead of on painstaking hand-drawn masks. Working with 476 field photos of banana bunches containing 1440 annotated defects, the authors find that training on SAM masks yields panoptic quality (PQ 0.779) identical to training on fully manual masks when both are evaluated against human ground truth, while cutting annotation workload by at least a factor of ten. They also show the trained model predicts the exact number of defects in 36.2% of images, is off by at most one in 76.2%, and estimates relative defect sizes with correlation r=0.96 against human annotation. The value of establishing this is practical: if it holds, defect-based grading of fresh produce becomes feasible in decentralized supply chains where large labeled datasets do not exist. The paper is candid about boundaries, reporting that SAM misses very small defects, over-segments long thin scars, and that four-class defect categorization is not reliable on this data.

What carries the argument

The machinery is a two-stage weak-supervision pipeline. First, SAM—the promptable Segment Anything Model—turns sparse prompts (bounding boxes around defects, click points on bananas) into dense instance masks without any training, with a ViT-L or ViT-H image encoder as the promptable backbone; this converts coarse annotations into pixel-level labels. Second, a MaskFormer model, pretrained on the ADE20k scene-parsing dataset, is trained on these SAM-generated masks to output panoptic predictions labeling Foreground Banana, Background Banana, and Defect instances. A postprocessing step then merges defect instances of the same class whose dilated masks overlap within a 2d-pixel distance (d=5), which notably improves defect AP. Finally, grading metrics are derived from the panoptic output: defect count from instance enumeration, and relative defect size from the ratio of defect pixels to foreground-banana pixels.

What would settle it

Photograph the same set of banana bunches with fixed physical defects from several distances and angles and check whether the relative defect-size ratio stays within the tolerance required for grading; if the ratio swings across viewpoints for the same physical defect, the size proxy fails and the reported r=0.96 does not translate into usable grading decisions.

Watch

Extended reading notes

Core claim

The central claim is that SAM-generated masks are a drop-in substitute for hand-annotated masks in training a MaskFormer panoptic segmentation model for banana surface defects. With the same postprocessing, the model trained on SAM-L masks scores PQ 0.779 against hand-annotated validation, exactly matching the PQ 0.779 of the model trained on fully annotated masks, while defect IoU drops only from 56.3% to 56.2% and defect AP from 22.3% to 21.7%. A larger SAM backbone (ViT-H) improves SAM-to-human mask alignment but does not improve downstream segmentation metrics, implying the training model absorbs some mask noise. The derived grading quantities are also close: exact defect count in 36.2% of images, within one defect in 76.2%, and a Pearson correlation of r=0.96 between predicted and annotated relative defect sizes across 793 paired defect instances. The four-way defect categorization (old/new bruise/scar) is reported as unreliable and is therefore not recommended on this dataset.

Load-bearing premise

The load-bearing premise is that relative defect size—defect pixels divided by foreground banana pixels in one uncalibrated 2D photo—faithfully reflects physical defect severity; without calibration, the same physical defect can look different at different distances, angles, and occlusions.

Editorial extensions

If this is right

  • Annotation effort for dense panoptic labels drops by at least a factor of 10: only boxes and points are needed, not pixel-wise masks.
  • A model trained on fewer than 500 field images with SAM-generated labels reaches the same panoptic quality as one trained on fully manual labels, so the approach is viable in low-data, decentralized settings.
  • The postprocessing merge step is a cheap, model-agnostic way to improve defect detection accuracy and can be applied to any semantic segmentation output.
  • Defect count and relative size, the two quantities most grading schemes need, are recoverable from panoptic masks with the reported accuracy.
  • The approach transfers, in principle, to other produce types whose grading depends on countable, sizable blemishes.

Reading between the lines

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

  • A natural next test is to calibrate the relative-size proxy against physical measurements (e.g., ruler or 3D scans) at multiple camera distances and angles; the paper does not provide that calibration, and the proxy is only faithful if viewpoint variation stays within grading tolerance.
  • SAM's failure below roughly 100 pixels of defect area implies a sensitivity floor: very small blemishes will be systematically missed, which sets a lower bound on what this pipeline can grade.
  • Because the paper uses a single annotator, the reported human ground truth itself contains ambiguity; multi-annotator agreement studies would show how much of the remaining error is model error versus annotation noise.
  • If the same weak-supervision recipe works for other commodities, the practical consequence is that grading pilots can be launched from a few hundred smartphone photos instead of large curated datasets.
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.

Referee Report

5 major / 6 minor

Summary. The paper proposes a weakly supervised pipeline for defect-based grading of fresh produce: SAM is prompted with coarse annotations (bounding boxes for defects, points for bananas) to generate dense panoptic masks, which are then used to train a Maskformer panoptic segmentation model. Experiments on 476 field images with 1,440 banana defects compare training on SAM-generated masks versus fully hand-annotated masks, evaluated against human-annotated ground truth. The central result is that both training label sources achieve nearly identical panoptic quality (PQ 0.779) on human validation. The paper also reports defect-count accuracy (36.2% exact, 76.2% within one) and a relative-size correlation of r=0.96 against a single expert's annotations, and it claims the annotation workload is reduced by at least a factor of 10.

Significance. If the central result holds, this is a practical contribution to low-data agricultural inspection: it demonstrates that foundation-model-generated masks can substitute for expensive manual mask annotation, with a sound experimental design that separates training and validation label sources. The paper is honest about failure cases (small defects, thin scars), includes a useful ablation of SAM backbone sizes, and releases code. However, the grading-utility claims (count and size accuracy) are currently supported only by agreement with a single annotator, the size correlation excludes non-matchable defects, the count and size metrics lack cross-validation variance, and the annotation-reduction factor is unmeasured. These issues do not undermine the central training comparison but need to be addressed before the broader claims are fully supported.

major comments (5)
  1. [Section 2.6 vs Section 3.3] The definition of relative defect size is inconsistent. Section 2.6 defines it as defect pixels divided by foreground banana pixels, whereas Section 3.3 states that the denominator is 'the aggregated number of pixels assigned to either the foreground banana class or any defect.' Please clarify which denominator was used to compute Figure 5 and reconcile the text, since the correlation value depends on this choice.
  2. [Section 3.3, Figure 5] The Pearson correlation r=0.96 is computed only on defect pairs matched with IoU >= 0.5, and non-matchable predicted defects are excluded. This selection can inflate the correlation. Report the fraction of predicted and annotated defects excluded, include an analysis that accounts for unmatched predictions (e.g., a matching-penalized metric or a sensitivity analysis at different IoU thresholds), and quantify the systematic underestimation visible from the regression slope deviating from the diagonal.
  3. [Section 3.3, Figure 4] The count accuracy percentages (36.2% exact, 76.2% within one) are reported without the cross-validation variance that is provided for Table 1. Please report these metrics per fold as mean and standard deviation, and specify which model configuration and postprocessing setting were used to produce Figure 4, so the reader can assess the robustness of the count claim.
  4. [Section 4] The statement that manual annotation workload is reduced 'by at least a factor of 10' is an estimate without supporting measurement. Provide annotation-time data for bounding-box/point prompts versus full mask annotation, or revise the claim to clearly indicate that it is an assumption. Since the abstract and discussion advertise this reduction, it should be substantiated or appropriately qualified.
  5. [Sections 2.2, 4, 5] All quantitative claims rest on annotations from a single expert, and the paper itself notes that defect boundaries and categories are often ambiguous. Without an inter-annotator agreement baseline, the reported counting and size accuracies cannot be interpreted as practical grading accuracy. The paper should present the count and size results as agreement with one expert rather than as absolute accuracy, and should state this limitation in the abstract and conclusion, not only in the future-work section.
minor comments (6)
  1. [Section 3.2, Table 1] The row with SAM-generated masks for both training and validation (PQ 0.782) is a self-consistency check rather than a direct comparison to human-validated rows. The text describing it as a 'slightly higher PQ score' may mislead; please clarify that the primary comparison for label-source equivalency is between the SAM-train/Anno-val row and the Anno-train/Anno-val row.
  2. [Section 2.5] The postprocessing step 3 states that instances within a distance of 2d pixels are merged, but this follows from a dilation by d pixels only implicitly. State explicitly that two instances are merged if their dilated masks overlap, i.e., if their original boundaries are within 2d pixels.
  3. [Figure 2] The x-axis bin labels are truncated in the rendered figure (e.g., '[10 , 1024²)' appears to be missing the exponent 4). Please update the figure or its caption so that all bin boundaries are legible.
  4. [Figure 4] The green and orange color coding in the count confusion matrix is not visible in grayscale reproductions. Add a legend or use distinct hatching to indicate the exact-match and within-one regions.
  5. [Section 3.3] The text reports n=793 matched defect pairs for the size analysis, but the dataset contains 1,440 annotated defects. Please define 'non-matchable' precisely and report how many annotated and predicted defects were excluded, so the reader can gauge the coverage of the correlation analysis.
  6. [Section 2.6] The pixel-ratio proxy for relative defect size is acknowledged as a simplification, but the paper does not discuss the varying camera distances and perspectives across the dataset. Add a sentence noting that these variations affect the pixel ratio and that the proxy is only meaningful for images taken under similar acquisition conditions.

Circularity Check

2 steps flagged · score 3.0 of 10

Central SAM-vs-hand annotation comparison is externally grounded, but SAM-SAM validation and SAM-assisted banana ground truth make some reported PQ/fruit-IoU numbers self-referential.

  1. fitted input called prediction [Section 3.2, Table 1 (rows 'Maskformer ✓ SAM-L SAM-L' and 'Maskformer ✓ SAM-H SAM-H')]
    "In the second experiment, we employ SAM-generated masks for both training and validation. The results show a slightly higher PQ score of 78.2% (ViT-L backbone)."

    The SAM-SAM rows use SAM-generated masks as both the training labels and the validation reference, so the resulting PQ measures the model's agreement with the same label generator that produced its training targets. Reporting 78.2% as 'slightly higher' than the 77.9% hand-annotated baseline compares different reference standards: the former is a self-consistency/fit score, statistically favored because the model is optimized toward SAM's label distribution, while the latter is an external human baseline. The separate SAM-train/human-val row remains a valid external comparison.

  2. fitted input called prediction [Section 2.2 (Banana annotations) and Table 1 fruit columns]
    "We annotated banana masks using LabelStudio and Segment Anything point prompts (see Section 2.3 for more details)."

    The foreground/background banana masks used as training and validation targets in every Table 1 row are dense masks produced by SAM from human point prompts—i.e., by the same foundation model used as the pseudo-label generator in the proposed pipeline. Consequently, the reported foreground-banana IoU (>95%) and the fruit component of PQ quantify agreement between MaskFormer and SAM's labeling, not agreement with fully manual human masks. Since the defect-mask comparison is validated against hand-annotated defect masks, the central claim is not reduced, but the absolute fruit and PQ numbers are partly self-referential.

full rationale

The paper's principal derivation chain is externally grounded: SAM generates dense defect masks from box/point prompts, MaskFormer is trained on these masks, and the key result is evaluated on held-out hand-annotated defect masks (SAM-train/Anno-val row, PQ 0.779 vs 0.779 for the hand-hand baseline). This comparison is not circular because the validation labels for defects come from independent human annotation. However, two reported results are partly self-referential. First, the SAM-SAM rows in Table 1 validate on masks generated by the same SAM process that created the training labels, so the PQ 78.2% is a self-consistency score presented alongside the human-baseline number. Second, the banana masks used for training and validation in all rows are themselves produced with SAM point prompts, so the high fruit IoU and the fruit component of PQ measure agreement with SAM's labeling rather than with fully manual masks. These issues do not collapse the central defect-mask comparison, but they mean that the absolute PQ/fruit numbers should be read as partially fitting the label generator. The paper also transparently flags the single-annotator ground truth and inter-annotator ambiguity (Section 5 and Appendix D), which is a validity concern about the human reference rather than an additional circularity.

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

The central claim rests on SAM's ability to turn coarse prompts into usable masks, MaskFormer's transfer from ADE20k, the reliability of a single expert's annotations, and the validity of the relative-size proxy. The pipeline itself introduces hand-chosen postprocessing parameters (d=5, IoU matching threshold 0.5) that directly shape the reported count and size accuracies. No fundamentally new entities are postulated.

free parameters (5)
  • Postprocessing merge distance d = 5 pixels
    Hand-chosen in Section 2.5 to merge defect instances whose expanded masks overlap (within 2d=10 pixels). The count accuracy depends on this value; no sensitivity analysis is provided.
  • IoU matching threshold for size evaluation = 0.5
    Section 3.3, Figure 5: predicted defects are paired with annotations only if IoU >= 0.5; non-matchable defects are excluded, which affects the reported r=0.96 size correlation.
  • Image resolution = 1024x1024
    All images resized and padded to 1024x1024 (Appendix B). Defects smaller than ~100 pixels are poorly captured; the resolution choice directly affects the small-defect failure mode.
  • Best-model selection = highest PQ on validation set
    Appendix B: model saved at epoch with highest validation PQ; this is a form of model selection on the validation set, standard practice but a hand choice.
  • SAM backbone for final pipeline = ViT-L
    The authors use SAM-L for the main results despite ViT-H giving higher mask IoU against human annotations (Section 3.2); they choose ViT-L for downstream training. This choice affects reported metrics.
assumptions (5)
  • domain assumption SAM's zero-shot mask generation from box and point prompts is a reliable proxy for dense manual annotation
    Section 2.3: the entire approach depends on SAM masks being good enough training labels; the paper evaluates this empirically but assumes it as the premise for the pipeline.
  • domain assumption MaskFormer pre-trained on ADE20k transfers to the banana defect domain
    Section 2.4: the panoptic model is initialized from ADE20k weights; no analysis of domain shift is provided.
  • domain assumption Single expert annotations constitute reliable ground truth for defect boundaries and categories
    Section 2.2: one annotator using LabelStudio; Section 4 acknowledges ambiguity, especially for old/new and bruise/scar categories; the detection evaluation assumes these labels are correct.
  • domain assumption Defect instances can be separated by connected components and dilation merging
    Section 2.5: the postprocessing assumes that same-class defects separated by more than 10 pixels are distinct instances; no alternative instance-separation validation is given.
  • standard math Standard definitions of IoU, AP, PQ, connected components, and dilation
    Used throughout Sections 2.4-2.6 and Appendix B; standard computer vision metrics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Weakly Supervised Panoptic Segmentation for Defect-Based Grading of Fresh Produce." pith.science (2026). https://pith.science/paper/FPF2R6MH

@misc{pith2026241116219,
  author       = {Pith},
  title        = {Pith review of: Weakly Supervised Panoptic Segmentation for Defect-Based Grading of Fresh Produce},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FPF2R6MH}},
  note         = {Machine review of arXiv:2411.16219}
}
read the original abstract

Visual inspection for defect grading in agricultural supply chains is crucial but traditionally labor-intensive and error-prone. Automated computer vision methods typically require extensively annotated datasets, which are often unavailable in decentralized supply chains. We address this challenge by evaluating the Segment Anything Model (SAM) to generate dense panoptic segmentation masks from sparse annotations. These dense predictions are then used to train a supervised panoptic segmentation model. Focusing on banana surface defects (bruises and scars), we validate our approach using 476 field images annotated with 1440 defects. While SAM-generated masks generally align with human annotations, substantially reducing annotation effort, we explicitly identify failure cases associated with specific defect sizes and shapes. Despite these limitations, our approach offers practical estimates of defect number and relative size from panoptic masks, underscoring the potential and current boundaries of foundation models for defect quantification in low-data agricultural scenarios. GitHub: https://github.com/manuelknott/banana-defect-segmentation

Figures

Figures reproduced from arXiv: 2411.16219 by the authors.

Figure 1
Figure 1. Overview of our approach. We utilize the Segment Anything Model (SAM), a promptable visual foundation model for image segmentation, to generate dense annotations (pixel-wise class and instance labels) from coarse annotations (bounding boxes and reference points) without any model training involved. These dense annotations otherwise require tedious hand-annotation. Using these newly generated labels, we train a panop… view at source ↗
Figure 2
Figure 2. Agreement between human-annotated and SAM￾predicted masks by defect size (number of pixels). The x-axis shows binned size categories for defect sizes in pixels (per anno￾tated mask), while n denotes the number of samples in each bin. The y-axis shows the agreement between annotated and predicted masks (IoU). SAM fails to align with human annotations for small (< 100 pixels) and very small (< 10 pixels) defects. To u… view at source ↗
Figure 3
Figure 3. SAM failure cases. Exemplary examples where SAM (ViT-L) fails to align with the annotated masks. We picked exam￾ples with the lowest IoU where the annotated mask size is > 100 pixels. The examples shown illustrate that long, thin scars are dif￾ficult to capture. Only one defect per image is shown. ment with human annotations across all defect sizes. This finding is confirmed by an additional analysis where we plot t… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Annotated versus predicted defect sizes. We pair￾wise match those defect instances with the highest IoU agreement between annotation and estimation (minimum 0.5) and calculate their sizes relative to the corresponding foreground banana masks (non-matchable defects are …
Figure 6
Figure 6. Figure 6: Example visualizations of annotated vs predicted masks. Left: Input Image, Mid: Annotation, Right: Maskformer Prediction. Red rectangles enclose defect instances. Segments are color-coded as follows: Foreground Banana , Background Banana , Defect , 7 [PITH_FULL_IMAGE:…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 27 canonical work pages

  1. [1]

    Automated sorting and grading of agricultural products based on image processing

    Hafiz Muhammad Tayyab Abbas, Usama Shakoor, Muhammad Jaleed Khan, Mushtaq Ahmed, and Khur- ram Khurshid. Automated sorting and grading of agricultural products based on image processing. In Proceedings of the 2019 8th International Confer- ence on Information and Communication Technolo- gies (ICICT), pages 78–81. Institute of Electrical and Electronics En...

    work page 2019

  2. [2]

    A Cookbook of Self-Supervised Learning

    Randall Balestriero, Mark Ibrahim, Vlad Sobal, Ari Morcos, Shashank Shekhar, Tom Goldstein, Florian Bordes, Adrien Bardes, Gregoire Mialon, Yuandong 8 Tian, Avi Schwarzschild, Andrew Gordon Wilson, Jonas Geiping, Quentin Garrido, Pierre Fernandez, Amir Bar, Hamed Pirsiavash, Yann LeCun, and Micah Goldblum. A Cookbook of Self-Supervised Learning. arXiv pre...

    arXiv 2023

  3. [3]

    New trends in the de- velopment and application of artificial intelligence in food processing

    Riya Barthwal, Deepika Kathuria, Saloni Joshi, RSS Kaler, and Narpinder Singh. New trends in the de- velopment and application of artificial intelligence in food processing. Innovative Food Science & Emerging Technologies, page 103600, 2024. Publisher: Elsevier. 1

    work page 2024

  4. [4]

    Schwing, and Alexander Kirillov

    Bowen Cheng, Alexander G. Schwing, and Alexander Kirillov. Per-Pixel Classification is Not All You Need for Semantic Segmentation. In NeurIPS, 2021. 4

    work page 2021

  5. [5]

    Schwing, Alexander Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, and Rohit Girdhar. Masked- attention Mask Transformer for Universal Image Seg- mentation. In CVPR, 2022. 4, 8

    work page 2022

  6. [6]

    Neill, and Artur Dubrawski

    Maria De-Arteaga, William Herlands, Daniel B. Neill, and Artur Dubrawski. Machine Learning for the De- veloping World. ACM Transactions on Management Information Systems, 9(2):1–14, 2018. 1

    work page 2018

  7. [7]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. arXiv preprint, 2020. 3

    work page 2020

  8. [8]

    Everingham, L

    M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2012 (VOC2012), 2012. 4

    work page 2012

Show all 33 references
  1. [9]

    Girshick

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross B. Girshick. Masked Autoen- coders Are Scalable Vision Learners.2022 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion (CVPR), pages 15979–15988, 2021. 3

    2022

  2. [10]

    What makes for effective detection pro- posals? IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(4):814–830, 2015

    Jan Hosang, Rodrigo Benenson, Piotr Dollár, and Bernt Schiele. What makes for effective detection pro- posals? IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(4):814–830, 2015. 4

    2015

  3. [11]

    A foundation model for cell segmentation

    Uriah Israel, Markus Marks, Rohit Dilip, Qilin Li, Changhua Yu, Emily Laubscher, Shenyi Li, Mor- gan Schwartz, Elora Pradhan, Ada Ates, Martin Abt, Caitlin Brown, Edward Pao, Alexander Pearson- Goulart, Pietro Perona, Georgia Gkioxari, Ross Barnowski, Yisong Yue, and David Van...

  4. [12]

    Adam: A method for stochastic optimization

    Diederik P Kingma. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. 11

    2014 arXiv

  5. [13]

    Girshick, Kaiming He, and Piotr Dollár

    Alexander Kirillov, Ross B. Girshick, Kaiming He, and Piotr Dollár. Panoptic Feature Pyramid Networks. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6392–6401, 2019. 4

    2019

  6. [14]

    Panoptic segmen- tation

    Alexander Kirillov, Kaiming He, Ross Girshick, Carsten Rother, and Piotr Dollár. Panoptic segmen- tation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9404–9413, 2019. 2, 4

    2019

  7. [15]

    Berg, Wan-Yen Lo, Piotr Dollár, and Ross Girshick

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollár, and Ross Girshick. Segment Anything. arXiv:2304.02643, 2023. 1, 2, 3

    2023 arXiv

  8. [16]

    Facilitated machine learning for image-based fruit quality assessment

    Manuel Knott, Fernando Perez-Cruz, and Thijs De- fraeye. Facilitated machine learning for image-based fruit quality assessment. Journal of Food Engineering, 345:111401, 2023. 1

    2023

  9. [17]

    Multimodal foundation models: From special- ists to general-purpose assistants

    Chunyuan Li, Zhe Gan, Zhengyuan Yang, Jianwei Yang, Linjie Li, Lijuan Wang, Jianfeng Gao, and oth- ers. Multimodal foundation models: From special- ists to general-purpose assistants. Foundations and Trends in Computer Graphics and Vision, 16(1-2):1– 214, 2024. Publisher: Now ...

    2024

  10. [18]

    Microsoft coco: Com- mon objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Com- mon objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzer- land, September 6-12, 2014, Proceed...

    2014

  11. [19]

    A visual-language foundation model for com- putational pathology

    Ming Y Lu, Bowen Chen, Drew FK Williamson, Richard J Chen, Ivy Liang, Tong Ding, Guillaume Jaume, Igor Odintsov, Long Phi Le, Georg Gerber, and others. A visual-language foundation model for com- putational pathology. Nature Medicine , 30(3):863– 874, 2024. Publisher: Nature P...

    2024

  12. [20]

    Open Source Computer Vision Library,

    OpenCV. Open Source Computer Vision Library,

  13. [21]

    SAM 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junting Pan, Kalyan Va- sudev Alwala, Nicolas Carion, Chao-Yuan Wu, Ross Girshick, Piotr Dollár, and Christoph Feichte...

    2024 arXiv

  14. [22]

    Vint: A foundation model for visual naviga- tion

    Dhruv Shah, Ajay Sridhar, Nitish Dashora, Kyle Sta- chowicz, Kevin Black, Noriaki Hirose, and Sergey Levine. Vint: A foundation model for visual naviga- tion. arXiv preprint arXiv:2306.14846, 2023. 1 9

    2023 arXiv

  15. [23]

    Get another label? improving data qual- ity and data mining using multiple, noisy labelers

    Victor S Sheng, Foster Provost, and Panagiotis G Ipeirotis. Get another label? improving data qual- ity and data mining using multiple, noisy labelers. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 614–622, 2008. 8

    2008

  16. [24]

    Label Studio: Data labeling software, 2020

    Maxim Tkachenko, Mikhail Malyuk, Andrey Hol- manyuk, and Nikolai Liubimov. Label Studio: Data labeling software, 2020. 3

    2020

  17. [25]

    The multidimensional wisdom of crowds

    Peter Welinder, Steve Branson, Pietro Perona, and Serge Belongie. The multidimensional wisdom of crowds. In Advances in Neural Information Process- ing Systems, 2010. 8

    2010

  18. [26]

    SegFormer: Simple and efficient design for semantic segmentation with transformers

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anand- kumar, Jose M Alvarez, and Ping Luo. SegFormer: Simple and efficient design for semantic segmentation with transformers. Advances in Neural Information Processing Systems, 34:12077–12090, 2021. 4

    2021

  19. [27]

    Depth any- thing: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth any- thing: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 10371–10381, 2024. 1

    2024

  20. [28]

    Implementation of information and communication technologies in fruit and vegetable supply chain: a systematic literature re- view

    M Yuvaraj, R Jothi Basu, Muhammad Dan-Asabe Ab- dulrahman, and C Ganesh Kumar. Implementation of information and communication technologies in fruit and vegetable supply chain: a systematic literature re- view. Industrial Management & Data Systems , 123 (9):2349–2377, 2023. Pu...

    2023

  21. [29]

    Text2seg: Re- mote sensing image semantic segmentation via text- guided visual foundation models

    Jielu Zhang, Zhongliang Zhou, Gengchen Mai, Lan Mu, Mengxuan Hu, and Sheng Li. Text2seg: Re- mote sensing image semantic segmentation via text- guided visual foundation models. arXiv preprint arXiv:2304.10597, 2023. 1

    2023

  22. [30]

    Scene Pars- ing through ADE20K Dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene Pars- ing through ADE20K Dataset. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5122–5130, 2017. 4 10 Supplementary Materials A. Dataset visualizations...

    2017

  23. [31]

    old” and

    Ambiguous annotation: The classification of defects across the four predefined categories may be subjective to a large degree. The distinction between “old” and ”new“ defects is not always clear-cut, as the transition is gradual. Additionally, distinguishing bruises from scars...

  24. [32]

    While this is generally considered high for machine learning tasks, defects can be small and, thus, only be represented by a few pixels, making them harder to categorize

    Image resolution: We use 10242 pixel resolution for images. While this is generally considered high for machine learning tasks, defects can be small and, thus, only be represented by a few pixels, making them harder to categorize

  25. [33]

    A more balanced distribution of categories and a larger number of defect samples are likely to improve categorization accuracy

    The four defect types are unevenly represented in our dataset (37/182/387/834). A more balanced distribution of categories and a larger number of defect samples are likely to improve categorization accuracy. We recommend collecting more examples from the two underrepresented c...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.