Pith. sign in

REVIEW 3 major objections 6 minor 36 references

Exploring Transfer Learning for Deep Learning Polyp Detection in Colonoscopy Images Using YOLOv8

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read The paper claims that pre-training on visually similar out-of-domain data beats in-domain medical data for polyp detection.

desk verdict Useful comparison of pre-training datasets for polyp detection, but the specific ranking claims rest on a leak-prone split and no error bars; worth refereeing if the authors tighten the statistics. read the letter →

arxiv 2502.00133 v1 pith:KKKOQVUU submitted 2025-01-31 cs.CV cs.AI

classification cs.CVcs.AI
keywords YOLOv8transferlearningpolypdetectioncolonoscopypre-trainingfine-tuningobjectdatasetselection
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

The paper asks which pre-training dataset gives the best transfer to polyp detection in colonoscopy images, and whether any pre-training beats training from scratch. It reports that a YOLOv8n model pre-trained on a combined fruit-and-vegetable dataset reached the highest F1-score (95.1%) after fine-tuning, outperforming models pre-trained on COCO, brain-tumor MRI, skin-lesion, and acne data. Across all epochs, every pre-trained model beat the from-scratch baseline, and models pre-trained on larger datasets beat their smaller counterparts. The authors conclude that shared visual features, such as round, shiny objects against textured backgrounds, matter more than the medical-domain label when choosing a pre-training set.

What carries the argument

The central object is the YOLOv8n object detector with transfer learning: a small (3.01 million parameter) network pre-trained on one of seven datasets and then fine-tuned on a combined polyp dataset. The transfer mechanism is weight initialization: the fine-tuning stage starts from weights learned on the pre-training dataset, and the same hyperparameters and augmentations are used across all runs to isolate the effect of pre-training choice. The comparison structure, pitting out-of-domain general datasets (COCO, fruit/vegetables) against in-domain medical sets (brain MRI, skin lesions, acne), is what carries the argument about which features transfer.

What would settle it

Compute the overlap of polyp identities between the training and test partitions of the combined dataset; if any polyp appears in both, re-evaluate all models on a patient-level split (or k-fold cross-validation) and check whether the fruit-and-veg model still tops YOLO-coco and whether every pre-trained model still beats from-scratch.

Watch

Extended reading notes

Core claim

The central claim is that transfer learning for polyp detection does not require an in-domain medical pre-training set; a dataset of fruits and vegetables, chosen for visual similarity to polyps, yields the strongest detector. The paper establishes this by pre-training YOLOv8n on seven datasets, fine-tuning on a combined set of four public polyp datasets, and evaluating F1, mAP50, and mAP50-95. The best model, YOLO-fruit&veg-XL, reached 95.1% F1 at 100 epochs, surpassing YOLO-coco (94.5%), all medical-image pre-trained models, and the from-scratch baseline (92.9%). The authors also find that pre-training on at least 10,000 images consistently improves results over smaller pre-training sets, and that every pre-trained model converges faster than the baseline. They interpret this as evidence that diversity and visual similarity are the driving factors, not the domain label itself.

Load-bearing premise

The single random 80/10/10 split assumes that images of the same polyp never appear in both the training and test sets; if they do, the reported F1 and mAP are inflated and the ranking between pre-training datasets could change.

Editorial extensions

If this is right

  • If replicated, polyp detection models can be built without large medical pre-training corpora, using publicly available food images instead.
  • Pre-training on larger, more diverse datasets (10k+ images) will consistently beat smaller pre-training sets, guiding dataset curation.
  • Any pre-training, even on acne or brain MRI, improves over from-scratch training, so practitioners should always use a pre-trained initialization.
  • The ranking suggests visual similarity (shape, shine, texture) should be a primary criterion for selecting pre-training data, not just domain closeness.

Reading between the lines

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

  • The paper's own limitation note says the fine-tuning set contains multiple images of the same polyp, so the single 80/10/10 split may leak polyp identity between training and test; a patient-level split could change the ranking, especially the small gaps between the top models.
  • The fruit-and-vegetable advantage hints that a specifically curated dataset of round, glossy, textured objects could be an even stronger pre-training set for polyp detection; that is testable.
  • The 'more data helps' result may conflate dataset size with dataset diversity; an experiment that controls class count while varying image count would separate the two effects.
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

3 major / 6 minor

Summary. The paper investigates whether pre-training YOLOv8n on different datasets improves polyp detection in colonoscopy images. It pre-trains on seven datasets (small: MRI brain tumor, acne, fruit/vegetable; larger XL: HAM10000, combined brain tumor, combined fruit/vegetable, plus a COCO-pretrained baseline), fine-tunes on a combined dataset of four public polyp datasets at 20/50/100 epochs, and reports precision, recall, F1, mAP50, and mAP50-95. The central claims are that pre-trained models consistently outperform training from scratch, that larger pre-training datasets help, and that the YOLO-fruit&veg-XL model achieves the highest overall F1-score of 95.1% after 100 epochs.

Significance. If validated, the paper would provide practical guidance on pre-training data selection for medical object detection and would contribute reusable pre-trained models and processed datasets. The experimental design is transparent and the authors explicitly release their code and models, which is a strength. However, the finer ranking claims rest on a single random split of a small dataset, with no error bars, no repeated seeds, and a split that may leak polyp identity between training and test; these issues must be addressed before the specific rankings can be trusted.

major comments (3)
  1. [§2.2 and §3.4] The fine-tuning set is split randomly at image level (80/10/10), while §3.4 acknowledges that the dataset 'consists of multiple images of the same polyp from different perspectives.' This means near-duplicate frames of the same polyp can appear in both training and test, which inflates reported F1/mAP and can change rankings. The differences between the top models are small (e.g., at 100 epochs fruit&veg-XL 95.1% vs. COCO 94.5% vs. HAM10000-XL 94.4%), so leakage of a few points is enough to alter the conclusions. Please stratify the split by polyp identity or video, or otherwise guarantee no identity leakage, and report results under a leakage-free protocol.
  2. [§3.2, Table 2] The conclusion that pre-trained models 'consistently outperformed' the scratch model 'at all epochs' is contradicted by Table 2: at 50 epochs, YOLO-acne (86.4% F1) and YOLO-brain-tumor (86.6% F1) fall below trained-from-scratch (88.9% F1), and the same is true for mAP50-95 (68.1% and 65.2% vs. 67.1%). Please narrow the claim to 'the best pre-trained models help' and explicitly discuss that some small-domain pre-training can hurt.
  3. [§3.2, Tables 4–6] The statement that YOLO-fruit&veg-XL is the best overall model is metric-dependent: although it has the highest F1 at 100 epochs (95.1%), HAM10000-XL has higher mAP50-95 (74.9% vs. 74.2%) and nearly equal mAP50 (97.0% vs. 97.1%). Moreover, all numbers come from a single random split with no repeated seeds or confidence intervals, so the 0.6-point F1 gap between fruit&veg-XL and COCO (95.1 vs. 94.5) cannot be distinguished from noise. Please add multiple seeds, report variance, and either apply statistical testing or explicitly state that the differences are not statistically significant.
minor comments (6)
  1. [§3.2] The first sentence of Section 3.2 is malformed: 'From all the results, the YOLO-fruit&veg-XL model achieved the highest overall F1-Among all the models evaluated...' should be rewritten as a single complete sentence.
  2. [Abstract] The abstract says 'objection tasks'; this should be 'object detection tasks'.
  3. [§2.3] The word 'treshold' appears twice and should be 'threshold'.
  4. [§3.1] The text says 'adam optimzer'; this should be 'Adam optimizer'.
  5. [§2.2] The text says 'The hyperparameters were kept consist'; this should be 'kept consistent'.
  6. [§3.4] The heading 'F uture W ork' has unintended spaces; also, 'k-means cross-validation' should be 'k-fold cross-validation'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reasoning: the paper reports measured transfer-learning comparisons; the central claims rest on empirical evaluations, not on assumptions that encode the conclusions.

full rationale

The paper is an empirical benchmark study: YOLOv8n models are pre-trained on seven datasets, fine-tuned on a combined polyp dataset, and evaluated on a held-out test split. The central claims (e.g., that models pre-trained on relevant or larger datasets outperform scratch training, and that YOLO-fruit&veg-XL reaches 95.1% F1) are directly supported by Tables 1-6, which report measured precision, recall, F1, and mAP values. No parameter is fitted to the outcome it is then said to predict: pre-training weights are learned on source datasets and then fine-tuned on the target polyp data, with the test split kept separate. There is no equation that defines the reported F1 or mAP in terms of the conclusion, and no self-citation carries the load of the argument; the cited prior works by the authors concern other medical-imaging tasks and are not used to justify the transfer-learning ranking. Concerns raised by the skeptical reading, such as the image-level 80/10/10 split possibly leaking multiple frames of the same polyp, the absence of multiple seeds, and the fact that Table 2 shows some small-domain pre-trained models (YOLO-acne 86.4, YOLO-brain-tumor 86.6) underperforming scratch (88.9) at 50 epochs, are genuine correctness and robustness issues, but they are not circularity: the results are measured, not derived from the assumptions. Similarly, the explanation that fruit/vegetable pre-training helps because of visual similarity is a post hoc interpretation, not a construction that forces the observed ranking. Under the criteria requiring a quotable reduction of a claim to its own inputs, no circular step exists, so the appropriate score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper's conclusions rest on three unstated data assumptions: annotation correctness of scraped public datasets, a fair comparison against Ultralytics' COCO model, and a split that avoids same-polyp leakage. No free parameters are fitted; all hyperparameters were fixed defaults.

assumptions (3)
  • domain assumption The 80/10/10 random split of the combined polyp datasets separates images of the same polyp.
    The authors note in Section 3.4 that the fine-tuning dataset consists of multiple images of the same polyp from different perspectives; if these leak across train and test, the evaluation is inflated and the ranking between pre-training datasets may change.
  • domain assumption The Ultralytics COCO-pretrained YOLOv8n, trained for 100 epochs, is directly comparable to the authors' 100-epoch pretraining runs.
    The COCO model is a third-party checkpoint; differences in augmentation, optimizer settings, or data pipelines between the Ultralytics run and the homemade pretraining could confound the comparison.
  • domain assumption The public Kaggle and Roboflow datasets contain accurate bounding boxes in YOLO format.
    The authors did not manually verify annotations on the scraped public datasets; noisy labels in pretraining would weaken feature transfer and could bias the comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Transfer Learning for Deep Learning Polyp Detection in Colonoscopy Images Using YOLOv8." pith.science (2026). https://pith.science/paper/KKKOQVUU

@misc{pith2026250200133,
  author       = {Pith},
  title        = {Pith review of: Exploring Transfer Learning for Deep Learning Polyp Detection in Colonoscopy Images Using YOLOv8},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KKKOQVUU}},
  note         = {Machine review of arXiv:2502.00133}
}
read the original abstract

Deep learning methods have demonstrated strong performance in objection tasks; however, their ability to learn domain-specific applications with limited training data remains a significant challenge. Transfer learning techniques address this issue by leveraging knowledge from pre-training on related datasets, enabling faster and more efficient learning for new tasks. Finding the right dataset for pre-training can play a critical role in determining the success of transfer learning and overall model performance. In this paper, we investigate the impact of pre-training a YOLOv8n model on seven distinct datasets, evaluating their effectiveness when transferred to the task of polyp detection. We compare whether large, general-purpose datasets with diverse objects outperform niche datasets with characteristics similar to polyps. In addition, we assess the influence of the size of the dataset on the efficacy of transfer learning. Experiments on the polyp datasets show that models pre-trained on relevant datasets consistently outperform those trained from scratch, highlighting the benefit of pre-training on datasets with shared domain-specific features.

Figures

Figures reproduced from arXiv: 2502.00133 by the authors.

Figure 1
Figure 1. Visual examples highlighting challenges in detecting polyps: (a) small polyps, (b) extremely large or abnormal [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overview of the transfer learning pipeline for deep learning-based polyp detection in colonoscopy images. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visual results of polyp detections by different methods. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    Global patterns and trends in colorectal cancer incidence and mortality,

    Arnold, M., Sierra, M. S., Laversanne, M., Soerjomataram, I., Jemal, A., and Bray, F., “Global patterns and trends in colorectal cancer incidence and mortality,” Gut 66(4), 683–691 (2017)

  2. [2]

    Early colorectal cancer detection with a spectroscopic liquid biopsy.,

    Cameron, J. M., Antoniou, G., Brennan, P., Conn, J., Gray, E., King, S., McHardy, R., Moug, S., Nobes, J., Palmer, D., et al., “Early colorectal cancer detection with a spectroscopic liquid biopsy.,” (2023)

  3. [3]

    Deep learning local- izes and identifies polyps in real time with 96% accuracy in screening colonoscopy,

    Urban, G., Tripathi, P., Alkayali, T., Mittal, M., Jalali, F., Karnes, W., and Baldi, P., “Deep learning local- izes and identifies polyps in real time with 96% accuracy in screening colonoscopy,”Gastroenterology155(4), 1069–1078 (2018)

  4. [4]

    Intracker: An integrated detector-tracker framework for cell detection and tracking,

    Liang, P., Chen, J., Zhang, Y., Wang, H., Zheng, H., Gu, P., and Chen, D., “Intracker: An integrated detector-tracker framework for cell detection and tracking,” in [ Processings of the IEEE 33rd International Symposium on Computer-Based Medical Systems], 332–337 (2020)

  5. [5]

    Boosting Medical Image Classification with Segmentation Foundation Model

    Gu, P., Zhao, Z., Wang, H., Peng, Y., Zhang, Y., Sapkota, N., Wang, C., and Chen, D. Z., “Boosting medical image classification with segmentation foundation model,” arXiv preprint arXiv:2406.11026(2024)

  6. [6]

    kCBAC-Net: Deeply supervised complete bipar- tite networks with asymmetric convolutions for medical image segmentation,

    Gu, P., Zheng, H., Zhang, Y., Wang, C., and Chen, D. Z., “kCBAC-Net: Deeply supervised complete bipar- tite networks with asymmetric convolutions for medical image segmentation,” in [ Processings of the Inter- national Conference on Medical Image Computing and Computer-Assisted Intervention], 337–347 (2021)

  7. [7]

    ConvFormer: Combining CNN and Transformer for medical image segmentation,

    Gu, P., Zhang, Y., Wang, C., and Chen, D. Z., “ConvFormer: Combining CNN and Transformer for medical image segmentation,” in [ Processings of the IEEE 20th International Symposium on Biomedical Imaging], 1–5 (2023)

  8. [8]

    SwIPE: Efficient and robust medical image segmen- tation with implicit patch embeddings,

    Zhang, Y., Gu, P., Sapkota, N., and Chen, D. Z., “SwIPE: Efficient and robust medical image segmen- tation with implicit patch embeddings,” in [ Processings of the International Conference on Medical Image Computing and Computer-Assisted Intervention], 315–326 (2023)

Show all 36 references
  1. [9]

    Sli2Vol+: Segmenting 3D medical images based on an object estimation guided correspondence flow network,

    An, D., Gu, P., Sonka, M., Wang, C., and Chen, D. Z., “Sli2Vol+: Segmenting 3D medical images based on an object estimation guided correspondence flow network,” arXiv preprint arXiv:2411.13873(2024)

  2. [10]

    You only look once: Unified, real-time object detection,

    Redmon, J., “You only look once: Unified, real-time object detection,” in [ Proceedings of the IEEE Confer- ence on Computer Vision and Pattern Recognition], (2016)

  3. [11]

    Colorectal polyp detection in colonoscopy images using yolo-v8 network,

    Lalinia, M. and Sahafi, A., “Colorectal polyp detection in colonoscopy images using yolo-v8 network,” Signal, Image and Video Processing18(3), 2047–2058 (2024)

  4. [12]

    WM-DOV A maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians,

    Bernal, J., S´ anchez, F. J., Fern´ andez-Esparrach, G., Gil, D., Rodr ´ ıguez, C., and Vilari˜ no, F., “WM-DOV A maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians,” Com- puterized Medical Imaging and Graphics43, 99–111 (2015)

  5. [13]

    Towards automatic polyp detection with a polyp appearance model,

    Bernal, J., S´ anchez, J., and Vilarino, F., “Towards automatic polyp detection with a polyp appearance model,” Pattern Recognition45(9), 3166–3182 (2012)

  6. [14]

    Toward embedded detection of polyps in wce images for early diagnosis of colorectal cancer,

    Silva, J., Histace, A., Romain, O., Dray, X., and Granado, B., “Toward embedded detection of polyps in wce images for early diagnosis of colorectal cancer,” International Journal of Computer Assisted Radiology and Surgery 9, 283–293 (2014)

  7. [15]

    Kvasir-seg: A segmented polyp dataset,

    Jha, D., Smedsrud, P. H., Riegler, M. A., Halvorsen, P., De Lange, T., Johansen, D., and Johansen, H. D., “Kvasir-seg: A segmented polyp dataset,” in [ Proceedings of the 26th International Conference of MultiMedia Modeling], 451–462 (2020)

  8. [16]

    Unbiased look at dataset bias,

    Torralba, A. and Efros, A. A., “Unbiased look at dataset bias,” in [ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition], 1521–1528 (2011)

  9. [17]

    Densely connected convolutional net- works,

    Huang, G., Liu, Z., Van Der Maaten, L., and Weinberger, K. Q., “Densely connected convolutional net- works,” in [ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition], 4700–4708 (2017)

  10. [18]

    A survey on transfer learning,

    Pan, S. J. and Yang, Q., “A survey on transfer learning,” IEEE Transactions on Knowledge and Data Engineering 22(10), 1345–1359 (2009)

  11. [19]

    Deep convolutional neural networks for computer-aided detection: Cnn architectures, dataset characteris- tics and transfer learning,

    Shin, H.-C., Roth, H. R., Gao, M., Lu, L., Xu, Z., Nogues, I., Yao, J., Mollura, D., and Summers, R. M., “Deep convolutional neural networks for computer-aided detection: Cnn architectures, dataset characteris- tics and transfer learning,” IEEE Transactions on Medical Imaging3...

  12. [20]

    Imagenet: A large-scale hierarchical image database,

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L., “Imagenet: A large-scale hierarchical image database,” in [ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition], 248–255 (2009)

  13. [21]

    Do better imagenet models transfer better?,

    Kornblith, S., Shlens, J., and Le, Q. V., “Do better imagenet models transfer better?,” in [ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition], 2661–2671 (2019)

  14. [22]

    Exploring deep learning and transfer learning for colonic polyp classification,

    Ribeiro, E., Uhl, A., Wimmer, G., and H¨ afner, M., “Exploring deep learning and transfer learning for colonic polyp classification,” Computational and Mathematical Methods in Medicine2016(1), 6584725 (2016)

  15. [23]

    Towards Alzheimer’s disease classification through transfer learning,

    Hon, M. and Khan, N. M., “Towards Alzheimer’s disease classification through transfer learning,” in [ Pro- ceedings of the the 2017 IEEE International Conference on Bioinformatics and Biomedicine], 1166–1169 (2017)

  16. [24]

    Transfer learning from a sparsely annotated dataset of 3D medical images,

    Humpire-Mamani, G. E., Jacobs, C., Prokop, M., van Ginneken, B., and Lessmann, N., “Transfer learning from a sparsely annotated dataset of 3D medical images,” arXiv preprint arXiv:2311.05032(2023)

  17. [25]

    Labeled mri brain tumor dataset

    Ahmed, A., “Labeled mri brain tumor dataset.” https://www.kaggle.com/datasets/ammarahmed310/ labeled-mri-brain-tumor-dataset (2023). Accessed: July 9, 2024

  18. [26]

    Acne dataset in yolov8 format

    Kurnaz, O. K., “Acne dataset in yolov8 format.” https://www.kaggle.com/datasets/osmankagankurnaz/ acne-dataset-in-yolov8-format (2023). Accessed: July 9, 2024

  19. [27]

    Fruits & vegetable detection for yolov4

    Patel, K., “Fruits & vegetable detection for yolov4.” https://www.kaggle.com/datasets/kvnpatel/ fruits-vegetable-detection-for-yolov4 (2023). Accessed: July 9, 2024

  20. [28]

    Microsoft coco: Common objects in context,

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll´ ar, P., and Zitnick, C. L., “Microsoft coco: Common objects in context,” in [ Proceedings of the European Conference on Computer Vision ], 740–755 (2014)

  21. [29]

    HAM1000 Segmentation and Classification

    Ghuwalewala, S., “HAM1000 Segmentation and Classification.” https://www.kaggle.com/datasets/ surajghuwalewala/ham1000-segmentation-and-classification (2018). Accessed: July 9, 2024

  22. [30]

    MRI for brain tumor with bounding boxes

    Sorour, A., “MRI for brain tumor with bounding boxes.” https://www.kaggle.com/datasets/ ahmedsorour1/mri-for-brain-tumor-with-bounding-boxes (2021). Accessed: July 9, 2024

  23. [31]

    Brain Tumor Dataset

    Nahin, M., “Brain Tumor Dataset.” https://www.kaggle.com/datasets/nahin333/ brain-tumor-dataset (2021). Accessed: July 9, 2024

  24. [32]

    Brain Tumor YOLO

    Pandey, A., “Brain Tumor YOLO.” https://www.kaggle.com/datasets/abhit007pandey/brain-tumor-yolo (2023). [Dataset]

  25. [33]

    Brian Tumor Dataset

    Fate52, “Brian Tumor Dataset.” https://universe.roboflow.com/fate52-l3uqx/brian-tumor/ dataset/2 (2024). Accessed: July 9, 2024

  26. [34]

    Fruit Detection

    Tyagi, L., “Fruit Detection.” https://www.kaggle.com/datasets/lakshaytyagi01/fruit-detection (2023). Accessed: July 9, 2024

  27. [35]

    Ultralytics yolov8,

    Jocher, G., Chaurasia, A., and Qiu, J., “Ultralytics yolov8,” (2023)

  28. [36]

    Adam: A method for stochastic optimization,

    Kingma, D. P. and Ba, J., “Adam: A method for stochastic optimization,” arXiv e-prints , arXiv–1412 (2014)

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.