The YOLO Framework: A Comprehensive Review of Evolution, Applications, and Benchmarks in Object Detection
Open accessThe DOI matches its Crossref record; 169 of the 169 references it cites were found in a registry or an index; 56 of the 154 sources its claims cite were checked for support.
Abstract
This paper provides a comprehensive review of the YOLO (You Only Look Once) framework up to its latest version, YOLO 11. As a state-of-the-art model for object detection, YOLO has revolutionized the field by achieving an optimal balance between speed and accuracy. The review traces the evolution of YOLO variants, highlighting key architectural improvements, performance benchmarks, and applications in domains such as healthcare, autonomous vehicles, and robotics. It also evaluates the framework’s strengths and limitations in practical scenarios, addressing challenges like small object detection, environmental variability, and computational constraints. By synthesizing findings from recent research, this work identifies critical gaps in the literature and outlines future directions to enhance YOLO’s adaptability, robustness, and integration into emerging technologies. This review provides researchers and practitioners with valuable insights to drive innovation in object detection and related applications.
Overview
CoChat Summary
TL;DR
This review traces the YOLO framework from early versions through YOLOv11, covering architectural changes, applications, benchmarks, and ethical concerns. It reports that YOLOv11 generally leads recent benchmarks, while YOLOv10 is close behind, and identifies persistent challenges in small-object detection, domain adaptability, computational cost, and responsible deployment.
Key findings
- YOLO has evolved through YOLOv1–YOLOv11, with key innovations including anchor boxes, CSPNet, PANet, PGI/GELAN in YOLOv9, C3k2 blocks in YOLOv10, and C2PSA/SPPF in YOLOv11.
- In the paper’s 20-epoch benchmark, YOLOv11 achieved the highest mAP50 across tested datasets: facemask 0.962, blood cell 0.958, forest smoke 0.945, human 0.854, pothole 0.815, and shellfish 0.563; YOLOv10 followed closely (e.g., 0.950, 0.944, 0.925, 0.829, 0.793, 0.542).
- YOLOv9 outperformed YOLO-NAS and YOLOv8 on complex detection tasks, but YOLOv10 and YOLOv11 further improved performance through better feature aggregation and spatial attention.
- The review documents applications in healthcare, autonomous vehicles, agriculture, and industry, and raises ethical concerns around privacy/surveillance, dataset bias, accountability, medical errors, environmental cost, and labor displacement.
Methods
- Comprehensive literature review using IEEE Xplore, SpringerLink, CVPR/ICCV/ECCV, Google Scholar, and other databases; initial pool of 53,200 articles, two-step title/abstract screening, refined to 135 full-text articles coded by architectural innovations, training strategies, performance metrics, and applications.
- Benchmark evaluation of YOLOv9, YOLO-NAS, YOLOv8, YOLOv10, and YOLOv11 on datasets including Roboflow 100, Object365, and COCO, with models trained for 20 epochs and mAP50 used as the primary metric.
Limitations
- Stated by the paper: benchmarks rely heavily on open-source datasets, limiting validation with real-world samples; the analysis focuses primarily on YOLO’s internal evolution rather than extensive comparison with other object detection frameworks; models were evaluated with a fixed number of training epochs, which may not capture optimal performance; and dataset biases may influence reported outcomes and generalizability.
- Evident from the text: the benchmark tables report mAP50 only, so speed, FLOPs, and resource-efficiency comparisons mentioned in the review’s coding scheme are not fully benchmarked in the presented results.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
49 Valid · 20 Not valid · 82 Not checked · 146 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“From detecting COVID-19 in X-ray images to enhancing road safety under adverse weather conditions [14], YOLO has demonstrated its ability to address complex challenges across fields such as medical imaging, autonomous vehicles, and agriculture.”
What the sources say
- Supportedcites [14] Ashraf, I., Hur, S., Kim, G., and Park, Y. (2024). Analyzing performance of YOLOx for detecting vehicles in bad weather conditions. Sensors, 24. source
The citation is attached to the adverse-weather road-safety example, not the examples of medical imaging or agriculture. The Abstract reports that the study tested YOLOx for vehicle detection in rain, fog, snow, and sandstorms; the Introduction places vehicle detection in the context of autonomous-vehicle safety. (LLM, conf=0.88)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“This capability makes YOLO exceptionally well-suited for applications requiring rapid decision-making, such as autonomous driving, medical diagnostics, and surveillance systems [8,9].”
What the sources say
- Partly supportedcites [8] Zhao, 2019, Object detection with deep learning: A review, IEEE Trans. Neural Netw. Learn. Syst. sourceExperimental — accuracy not yet measured
The YOLO discussion (p. 9) reports real-time processing at 45 FPS, supporting its suitability where speed matters. The paper mentions autonomous driving and surveillance as object-detection applications (Introduction, p. 1; Future Directions, p. 17), but does not establish that YOLO specifically is exceptionally well-suited to those settings or to medical diagnostics. (LLM, conf=0.85)
- Partly supportedcites [9] Ragab, 2024, A Comprehensive Systematic Review of YOLO for Medical Object Detection (2018 to 2023), IEEE Access sourceExperimental — accuracy not yet measured
Our copy supports the speed/real-time basis of the claim and its medical-diagnostics instance: the Advantages section (p. 6) states YOLO's single-pass ability 'enables real time processing of medical images, which is crucial in time-sensitive medical scenarios such as surgeries or emergency diagnostics', and Medical Imaging (p. 9) says 'YOLO's real time performance makes it wellsuited for tasks such as identifying fractures in X-rays'. However, Ragab et al. is a review scoped to medical object detection (Introduction, p. 1: applications in medical imaging, surgical procedures, personal protective equipment detection); no body passage states that YOLO is well-suited to autonomous driving or to surveillance systems. Autonomous driving appears only as a reference entry ([133] Andrade et al., References p. 24), and 'surveillance' is not treated as an application domain — the closest is patient-monitoring/safety video feeds in healthcare (p. 14). So the two non-medical exemplars in the citing sentence are not attributable to this source. (LLM, conf=0.78)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“YOLOv3 has been used to detect red lesions in RGB fundus images, achieving an average precision of 83.3% [55].”
What the sources say
- Partly supportedcites [55] Monemian, M., and Rabbani, H. (2023). Detecting red-lesions from retinal fundus images using unique morphological features. Sci. Rep., 13. sourceExperimental — accuracy not yet measured
The References identify a YOLO V3 study of red-lesion detection, and Experimental results describes its CNN method. But Experimental results and Table 2 report 83% sensitivity for that study, not 83.3% average precision; the cited passages also do not specify RGB images. (LLM, conf=0.88)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“YOLOv3, paired with data augmentation techniques like 180 ◦ and 90 ◦ rotations, has proven effective in brain tumor detection and segmentation.”
What the sources say
- Partly supportedcites [62] Hossain, 2021, A YOLOv3 Deep Neural Network Model to Detect Brain Tumor in Portable Electromagnetic Imaging System, IEEE Access sourceExperimental — accuracy not yet measured
The augmentation half of the claim is supported: Sec. IV.A 'Procedure of Training, Validation and Testing' lists eight augmentation techniques including rotation at angle ranges '-90 to 90 degrees' and '-180 to 180 degrees', applied to 50 sample images to build a 1000-image data set, and the Abstract/Sec. V report detection accuracy 95.62% and F1 94.50%, i.e. YOLOv3 was effective at tumor detection (in the portable EM imaging system with a tissue-mimicking head phantom). However, nothing in our copy supports 'and segmentation': the model outputs regression bounding boxes ('predicted output with bounding boxes is displayed as a final target object detection result', Step 4) and the Conclusion frames the contribution as 'detect and localize the brain tumor', with no pixel-level or tumor-mask segmentation task or metric; segmentation appears only in titles of cited works (refs 38-39). (LLM, conf=0.82)
- Partly supportedcites [63] Chen, A., Lin, D., and Gao, Q. (2024). Enhancing brain tumor detection in MRI images using YOLO-NeuroBoost model. Front. Neurol., 15. sourceExperimental — accuracy not yet measured
Section 2.2 (page not available) says YOLOv3 improved the speed and accuracy of brain tumor detection in MRI images. It does not link YOLOv3 to 180° or 90° rotations or say it was effective for segmentation; Section 3.2 discusses data augmentation for YOLOv8 instead. (LLM, conf=0.88)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“YOLOv7 has been utilized to compress medical images while maintaining high-quality reconstructions during transmission [64].”
What the sources say
- Could not verifycites [64] Sara, 2019, Image Quality Assessment through FSIM, SSIM, MSE and PSNR—A Comparative Study, J. Comput. Commun. sourceExperimental — accuracy not yet measured
The Introduction discusses image quality during compression and transmission, and the PSNR discussion describes assessing reconstruction quality after lossy compression. Neither passage addresses YOLOv7 or the compression of medical images; searches for those terms found no supporting passage in our copy. (LLM, conf=0.76)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
108 more claims · 16 with their cited sources checked
Create a free account to see all 297 claims and to verify the 50 references still open.
- It may state two separate claims as one. 113
- Probably not a substantial claim. 106
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 5.1.3. Evolution to YOLOv10 and YOLOv11
“A key advancement in YOLOv10 was the introduction of the C3k2 block, an innovative feature that greatly improved feature aggregation while reducing computational overhead [138].”
§ 2.3. Coding and Classification
“| Introduced C3k2 block in backbone and used C2PSA to enhance spatial attention | Fine-tuning, mix-up, augmentation, adaptive gradient clipping | Quantization, stochastic gradient descent (SGD) | YOLOV11”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 3.2.4. RFB Net
“The RFB Net (RefineNet with Anchor Boxes) [34], developed by Niu and Zhang, builds upon the basic RefineNet architecture.”
§ References
“34. Liu, S.; Huang, D. Receptive field block net for accurate and fast object detection.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.4.1. Applications of YOLO in Industrial Manufacturing
“Workpiece Detection and Localization: Another notable application is workpiece detection on production lines, where YOLOv5 was enhanced with lightweight deep convolution layers to improve detection accuracy. The model showed significant improvements, increasing mAP by 2.4% on the COCO dataset and 4.2% on custom industrial datasets [128].”
§ References
“128. Xu, X.; Zhang, X.; Zhang, T. Lite-YOLOv5: A Lightweight Deep Learning Detector for On-Board Ship Detection in Large-Scene Sentinel-1 SAR Images.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.2.1. Benchmark Findings
“Our results consistently showed that YOLOv9 outperformed both YOLO-NAS and YOLOv8, particularly in complex image detection tasks where objects are occluded or exhibit detailed patterns.”
§ 5.2.1. Benchmark Findings
“| Forest Smoke Detection | 0.865 | 0.7811 | 0.911 | 0.925 | 0.945 | | Human Detection | 0.812 | 0.7145 | 0.816 | 0.829 | 0.854 |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.2.4. Efficiency and Speed Optimizations for Autonomous Vehicles
“For instance, YOLOv7 has been enhanced with a hybrid attention mechanism (ACmix) and the Res3Unit backbone, significantly improving its performance, achieving an AP score of 90.8% on road traffic data [80].”
§ 4.2.5. YOLO's Performance Across Different Lighting Conditions
“| YOLOv7 [80] | AP | Proposed modifications including ACmix for hybrid attention, RFLA for feature fusion, and Res3Unit backbone, achieving 89.2% AP. | Road traffic data |”
2 more findings
Create a free account to see every finding of this check.
Cited works that were retracted
Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.
A note, not an issue: for example, a cited work that was retracted only after this paper was published, which its authors could not have known.
In the paper
“This capability is essential to ensure the safety and reliability of autonomous driving systems.”
Cited work 10.3390/diagnostics12040823 · retracted 2026-08-26 · notice 10.3390/diagnostics16172728 · this paper published 2024-12-14
Citations
Cites
1 retracted
Amin, J., Anjum, M., Sharif, M., Kadry, S., Nadeem, A., & Ahmad, S. (2022) · Diagnostics · doi:10.3390/diagnostics12040823 · Cited by 38 · cited 1×
“This capability is essential to ensure the safety and reliability of autonomous driving systems.”
RetractedFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812
You Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460
(2016) · Cited by 31,590
Focal Loss for Dense Object Detection
Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollar, P. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.324 · Cited by 30,581
SSD: Single Shot MultiBox Detector
Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46448-0_2 · Cited by 22,128
EfficientDet: Scalable and Efficient Object Detection
Tan, M., Pang, R., & Le, Q. V. (2019) · arXiv · Cited by 9,841 · cited 2×
“Detailed work fl ow of RetinaNet.”
YOLOv10: Real-Time End-to-End Object DetectionPreprint
Wang, A., Chen, H., Liu, L., et al. (2024) · Neural Information Processing Systems · doi:10.48550/arxiv.2405.14458 · Cited by 6,192
Research works printed but not linked: 1
- Wang, X., Li, H., Yue, X., and Meng, L. (2023, January 28–29). A comprehensive survey on object detection YOLO. Proceedings of the 5th International Symposium on Advanced Technologies and Applications in the Inter of Things (ATAIT), Kusatsu, Japan.
Cited by
A comprehensive review on YOLO versions for object detection
Murat, A. A., & Kıran, M. S. (2025) · Engineering Science and Technology an International Journal · doi:10.1016/j.jestch.2025.102161 · Cited by 228
Rose-Mamba-YOLO: an enhanced framework for efficient and accurate greenhouse rose monitoring
You, S., Li, B., Chen, Y., Ren, Z., Liu, Y., Wu, Q., Tao, J., Zhang, Z., Zhang, C., Xue, F., et al. (2025) · Frontiers in Plant Science · doi:10.3389/fpls.2025.1607582 · Cited by 65
Self-citationRP-DETR: end-to-end rice pests detection using a transformer
Wang, J., Wang, T., Xu, Q., Gao, L., Gu, G., Jia, L., & Yao, C. (2025) · Plant Methods · doi:10.1186/s13007-025-01381-w · Cited by 22
Yang, X., Shao, S., Liu, W., Xiang, Z., & Zhang, B. (2025) · Ocean Engineering · doi:10.1016/j.oceaneng.2025.122618 · Cited by 27
Real-time safety detection on construction sites using a vision-language and NLP-based model
Sivanraj, S., Uduwage, D. N. L. S., & Tripathi, M. (2025) · Advanced Engineering Informatics · doi:10.1016/j.aei.2025.103889 · Cited by 30
Metadata
- Authors
- Ali, M. L., & Zhang, Z.
- Affiliations
- Middle Tennessee State University (US)
- Farmingdale State College (US)
- Corresponding author
- Zhang, Zhou
- Venue
- Computers
- Publisher
- MDPI AG
- Published
- 2024-12-14
- Volume
- 13
- Issue
- 12
- Pages
- 336
- ISSN
- 2073-431X (ISSN-L)
- OpenAlex
- W4405446156
- Cited by
- 383
- References
- 151
- Citations by year
- 2026: 242 · 2025: 246 · 2024: 3
- FWCI
- 55.91
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Computers (publishedVersion, cc-by)
- Preprints.org (acceptedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
- Keywords
- YOLO, object detection, real-time detection, small object detection, autonomous vehicles, robotics, medical applications, benchmark performance
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.