Full search page →

YOLO26: a DFL-free end-to-end YOLO

Optional · paper guide 16 of 22 for DLCV 03A · One-Stage Detection, Read Through YOLOv4. Citation numbers refer to the reference list at the bottom of this page.

Navigate: this page and this track: Navigate: this page and this track

On this page

Supplementary notes · 2

  1. Ultralytics releases without a paper
  2. Folding a re-parameterized block into one convolution

Paper guides · 22

  1. YOLO v1: detection as a single regression
  2. YOLOv2 and YOLO9000: anchors, priors and joint training
  3. YOLOv3: three scales and Darknet-53
  4. SSD: multi-scale default boxes
  5. RetinaNet and the focal loss
  6. FCOS: per-pixel anchor-free detection
  7. CenterNet: objects as points
  8. YOLOR: explicit features and learned implicit representations
  9. YOLOX: anchor-free YOLO with SimOTA
  10. YOLOv6: a re-parameterized industrial detector
  11. YOLOv7: a trainable bag of freebies
  12. YOLOv9: programmable gradient information
  13. YOLOv10: NMS-free training with dual assignments
  14. YOLOv12: an attention-centric YOLO
  15. YOLOv13: hypergraph correlation
  16. YOLO26: a DFL-free end-to-end YOLO
  17. RT-DETR: a real-time DETR
  18. RT-DETRv2: a bag of freebies for RT-DETR
  19. D-FINE: boxes as refined distributions
  20. DEIM: dense one-to-one matching
  21. RF-DETR: architecture search on a DINOv2 backbone
  22. DEIMv2: DINOv3 features for real-time DETRs

Deep Learning for Computer Vision

  1. 02 · Deep Image Classification
View the whole course tree →

Paper: G. Jocher, J. Qiu, M. Liu, S. Lyu, F. C. Akyon and M. E. Kalfaoglu (Ultralytics), “Ultralytics YOLO26: Unified real-time end-to-end vision models,” arXiv, June 2026 [20]; the software was released in January 2026. Related part of YOLOv4: label assignment, post-processing and the training schedule.

Challenge

The paper lists four limits of current YOLOs [20]: (a) most still need NMS, and YOLOv10’s dual heads give both heads the same loss weight throughout training, leaving the inference head under-trained; (b) the distribution focal loss (DFL), used in the Ultralytics head since YOLOv8, expands each box side from one number to K=16K = 16 logits, inflating the head (YOLO11n has 2.6 M parameters and 6.5 GFLOPs with DFL and 2.3 M and 5.2 GFLOPs without), and limits the range of each side; (c) training needs about 600 epochs; (d) task-aligned assignment selects candidates whose anchor centres lie inside the box, so tiny objects can receive no positive at all.

Method

Read the training pipeline from shared features through dual heads and assignment to progressive loss weights and MuSGD. The two training losses do not mean both heads are required in default inference.
Original paper figure 2 — Read the training pipeline from shared features through dual heads and assignment to progressive loss weights and MuSGD. The two training losses do not mean both heads are required in default inference.

Figure source: [20], arXiv:2606.03748v1, PDF page 6. Original PDF.

Dual heads, NMS-free by default. As in YOLOv10, a one-to-one head and a one-to-many head share the backbone and neck and use the same task-aligned assignment with different cardinalities: the one-to-one path forms a candidate set with top-k=7k = 7 and then keeps the top 1 per object; the one-to-many head uses top-k=10k = 10 [20]. The one-to-one head outputs at most 300 detections per image without NMS; the one-to-many head remains available with NMS when the highest accuracy matters [20].

No DFL. DFL [80] predicts a discrete distribution over KK bins for each box side and decodes the side as its expected value d=∑i=0K−1i⋅softmax(z)id = \sum_{i=0}^{K-1} i \cdot \text{softmax}(z)_i, so a side can be at most (K−1)(K - 1) strides long: about 960 pixels of width at stride 32 with K=16K = 16 [20]. YOLO26 regresses the box directly with an L1 loss and recovers the lost localization quality with the training changes below [20]. The DFL loss itself, as defined in GFL, is

y^=∑i=0nSi yi,DFL(Si,Si+1)=−((yi+1−y)log⁡Si+(y−yi)log⁡Si+1),\hat{y} = \sum_{i=0}^{n} S_i\, y_i, \qquad \text{DFL}(S_i, S_{i+1}) = -\big((y_{i+1} - y)\log S_i + (y - y_i)\log S_{i+1}\big),

where SiS_i is the softmax probability of the bin value yiy_i, yy the true distance and yi≤y≤yi+1y_i \le y \le y_{i+1} its two neighbouring bins; the minimum puts the mass on the two bracketing bins in proportion to the distances, so that y^=y\hat{y} = y exactly [80]. The snippet shows the decoding, the loss and the range limit.

Code · Python · 22 lines
import numpy as np

K, stride = 16, 8                        # 16 bins per box side, a stride-8 level
bins = np.arange(K)                      # support y_0 = 0, ..., y_15 = 15 (in stride units)

def softmax(z):
    e = np.exp(z - z.max())
    return e / e.sum()

def dfl(logits, y):
    """Distribution focal loss for one box side with continuous target y (stride units)."""
    S = softmax(logits)
    i = int(np.floor(y))                 # neighbours y_i <= y <= y_i + 1
    return -((i + 1 - y) * np.log(S[i]) + (y - i) * np.log(S[i + 1]))

y = 5.3                                  # true distance: 5.3 cells = 42.4 pixels
flat = np.zeros(K)                       # untrained head: uniform distribution
sharp = np.full(K, -10.0); sharp[5], sharp[6] = np.log(0.7), np.log(0.3)   # the ideal shape
for name, z in [("uniform", flat), ("ideal", sharp)]:
    S = softmax(z)
    print(f"{name:7s} expected distance {S @ bins:5.2f} cells = {S @ bins * stride:6.1f} px | DFL {dfl(z, y):.3f}")
print("largest distance this head can output:", (K - 1) * stride, "px")

Output

uniform expected distance  7.50 cells =   60.0 px | DFL 2.773
ideal   expected distance  5.30 cells =   42.4 px | DFL 0.611
largest distance this head can output: 120 px

The ideal distribution decodes to exactly 5.3 cells, but its loss is not zero (0.611 is the entropy of the 0.7 / 0.3 split), and the support is finite: with 16 bins a side can be at most 15 strides long. Figure 6 shows the three cases.

Three bar charts over 16 bins. Left: a flat distribution whose expected value is 7.50, far from the true distance 5.3. Middle: probability 0.7 at bin 5 and 0.3 at bin 6, whose expected value is exactly 5.30. Right: a broad bell-shaped distribution centred near bin 6 with expected value 6.01.
Figure 6 — How a distribution-based box side is decoded. The dashed line is the expected value (the predicted distance); the yellow line is the true distance 5.3. DFL trains the head toward the middle panel; a broad distribution, as on the right, expresses uncertainty about the edge.

Progressive loss. The total loss shifts weight from the one-to-many to the one-to-one head during training:

Ltotal=α(t) Lone2many+(1−α(t)) Lone2one,α(t)=max⁡ ⁣(1−tmax⁡(E−1,1), 0)(αinit−αfinal)+αfinal,\mathcal{L}_{\text{total}} = \alpha(t)\, \mathcal{L}_{\text{one2many}} + \big(1 - \alpha(t)\big)\, \mathcal{L}_{\text{one2one}}, \qquad \alpha(t) = \max\!\Big(1 - \frac{t}{\max(E - 1, 1)},\, 0\Big)(\alpha_{\text{init}} - \alpha_{\text{final}}) + \alpha_{\text{final}},

where tt is the current epoch, EE the number of epochs and αinit,αfinal\alpha_{\text{init}}, \alpha_{\text{final}} the initial and final one-to-many weights [20]. The default starts at weights (0.8, 0.2) for (one-to-many, one-to-one) and ends at (0.1, 0.9) [20].

STAL (small-target-aware label assignment). For candidate selection only, every box side shorter than the smallest stride smin⁡s_{\min} is enlarged to a reference size srefs_{\text{ref}} (the second pyramid stride), so that at least one anchor centre falls inside a tiny box; the original box is kept for scoring, assignment and regression [20].

MuSGD. A hybrid optimizer: for multi-dimensional weights it mixes the SGD-momentum update with the Muon update [86], which orthogonalizes the momentum-based update with a few Newton–Schulz iterations; biases and normalization parameters use plain SGD [20].

The paper also adds task-specific heads for segmentation, pose and oriented boxes, and an open-vocabulary variant; these are outside this chapter.

Results

Table 7 lists the released detection models on COCO val at 640 pixels, with T4 TensorRT 10 latency and CPU ONNX latency [20]: YOLO26n reaches 40.9% AP with NMS on the one-to-many head and 40.1% end to end at 1.7 ms; YOLO26s 48.6% / 47.8% at 2.5 ms; YOLO26x 57.5% / 56.9% at 11.8 ms. The end-to-end head trails the NMS head by 0.6 to 0.8 AP across sizes [20]. The released models are pre-trained on Objects365 [87] before COCO fine-tuning [20], so they are not directly comparable with COCO-only models.

Ablations

Table 2 goes from YOLO11s to YOLO26s, reporting the NMS-based AP and, from the end-to-end step on, the end-to-end AP [20]: YOLO11s 47.0%; removing DFL 46.4%; + L1 loss 46.6%; + STAL 46.8%; + backbone and neck refinement 47.0%; + end-to-end head 46.4% end to end (47.0% with NMS); + progressive loss 46.7% (47.2%); + MuSGD 47.1% (47.6%); + Objects365 pre-training 47.4% (48.0%); + hyper-parameter search 47.8% (48.6%). A controlled DFL ablation at 640 and 1280 pixels finds that the DFL-free head is better at both, by 0.3 AP (1.0 APL) at 640 and 1.3 AP (2.2 APL) at 1280 [20]. MuSGD reaches 47.4% AP in 500 epochs against 47.0% for SGD in 600 [20]. For the progressive loss, starting at (1.0, 0.0), which gives the one-to-one head no weight at first, does not help (46.3% to 46.4% end to end), while the default schedule reaches 46.7% [20]. For STAL, sref=16s_{\text{ref}} = 16 is best (46.8% AP, small-object AP 29.6% against 29.0%), while 8 and 32 match or underperform the baseline [20].

Conclusion

YOLO26 makes the NMS-free head the default deployment path, removes the distribution-based box head and compensates with training changes aimed at the inference head and at tiny objects, ending where YOLOv4’s anchor-based, NMS-dependent design started from the opposite side.

Problems

It is an arXiv paper by the company that ships the software, with benchmarks from its own repository; its main numbers use Objects365 pre-training and a hyper-parameter search; each training component adds a few tenths of AP in single runs; and the end-to-end head is still 0.6 to 0.8 AP below the NMS head. The paper itself names broader evaluation beyond COCO-centric benchmarks and learned loss schedules as future work [20].

References

  1. C.-Y. Wang, A. Bochkovskiy and H.-Y. M. Liao, “YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,” in Proc. IEEE/CVF CVPR, pp. 7464–7475, 2023. doi
  2. C.-Y. Wang, I-H. Yeh and H.-Y. M. Liao, “YOLOv9: Learning what you want to learn using programmable gradient information,” in Proc. ECCV, 2024. doi
  3. G. Jocher, “Ultralytics YOLOv5,” software (first released 2020; v7.0 released 22 November 2022). release · documentation
  4. G. Jocher, A. Chaurasia and J. Qiu, “Ultralytics YOLOv8,” software, version 8.0.0, 2023 (no paper). docs · code
  5. G. Jocher and J. Qiu, “Ultralytics YOLO11,” software, version 11.0.0, 2024 (no paper). docs · code
  6. J. Lu, J. Jia, J. Yawl and C. Zhang, “YOLOv14: Adaptive real-time object detection for diverse imaging conditions,” arXiv:2608.04720, 2026 (withdrawn by the authors, 20 August 2026). arXiv
  7. J. Redmon, S. Divvala, R. Girshick and A. Farhadi, “You only look once: Unified, real-time object detection,” in Proc. IEEE CVPR, pp. 779–788, 2016. doi · arXiv
  8. J. Redmon and A. Farhadi, “YOLO9000: Better, faster, stronger,” in Proc. IEEE CVPR, pp. 6517–6525, 2017. doi · arXiv
  9. J. Redmon and A. Farhadi, “YOLOv3: An incremental improvement,” tech. report, arXiv:1804.02767, 2018. arXiv
  10. W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C. Fu, A. C. Berg, “SSD: Single Shot MultiBox Detector,” Lecture Notes in Computer Science / Computer Vision – ECCV 2016, 2016. doi · manuscript
  11. T.-Y. Lin, P. Goyal, R. Girshick, K. He and P. Dollár, “Focal loss for dense object detection,” in Proc. IEEE ICCV, pp. 2999–3007, 2017. doi
  12. Z. Tian, C. Shen, H. Chen, T. He, “FCOS: Fully Convolutional One-Stage Object Detection,” 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019. doi · manuscript
  13. X. Zhou, D. Wang, P. Krähenbühl, “Objects as Points,” arXiv preprint, 2019. source
  14. C. Wang, I. Yeh, H. M. Liao, “You Only Learn One Representation: Unified Network for Multiple Tasks,” arXiv preprint, 2021. source
  15. Z. Ge, S. Liu, F. Wang, Z. Li and J. Sun, “YOLOX: Exceeding YOLO series in 2021,” arXiv:2107.08430, 2021. arXiv
  16. C. Li, L. Li, H. Jiang, K. Weng, Y. Geng et al., “YOLOv6: A single-stage object detection framework for industrial applications,” tech. report, arXiv:2209.02976, 2022. arXiv
  17. A. Wang, H. Chen, L. Liu, K. Chen, Z. Lin, J. Han and G. Ding, “YOLOv10: Real-time end-to-end object detection,” in Proc. NeurIPS, 2024. arXiv
  18. Y. Tian, Q. Ye and D. Doermann, “YOLOv12: Attention-centric real-time object detectors,” in Advances in Neural Information Processing Systems 38 (NeurIPS), 2025. proceedings · arXiv
  19. M. Lei, S. Li, Y. Wu, H. Hu, Y. Zhou et al., “YOLOv13: Real-time object detection with hypergraph-enhanced adaptive visual perception,” arXiv:2506.17733, 2025. arXiv
  20. G. Jocher, J. Qiu, M. Liu, S. Lyu, F. C. Akyon and M. E. Kalfaoglu, “Ultralytics YOLO26: Unified real-time end-to-end vision models,” arXiv:2606.03748, 2026 (software released January 2026). arXiv · docs
  21. Y. Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, Y. Liu, J. Chen, “DETRs Beat YOLOs on Real-time Object Detection,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024. source
  22. W. Lv, Y. Zhao, Q. Chang, K. Huang, G. Wang and Y. Liu, “RT-DETRv2: Improved baseline with bag-of-freebies for real-time detection transformer,” arXiv:2407.17140, 2024. arXiv
  23. Y. Peng, H. Li, P. Wu, Y. Zhang, X. Sun and F. Wu, “D-FINE: Redefine regression task in DETRs as fine-grained distribution refinement,” in Proc. ICLR, 2025. arXiv · code
  24. S. Huang, Z. Lu, X. Cun, Y. Yu, X. Zhou, X. Shen, “DEIM: DETR with Improved Matching for Fast Convergence,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. source
  25. I. Robinson, P. Robicheaux, M. Popov, D. Ramanan and N. Peri, “RF-DETR: Neural architecture search for real-time detection transformers,” in Proc. ICLR, 2026. arXiv
  26. S. Huang, Y. Hou, L. Liu, X. Yu and X. Shen, “Real-time object detection meets DINOv3,” arXiv:2509.20787, 2025 (revised 2026). arXiv
  27. A. Bochkovskiy, C.-Y. Wang and H.-Y. M. Liao, “YOLOv4: Optimal speed and accuracy of object detection,” arXiv:2004.10934, 2020. arXiv
  28. M. Tan, R. Pang, Q. V. Le, “EfficientDet: Scalable and Efficient Object Detection,” 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020. doi · manuscript
  29. NVIDIA, “TensorRT documentation,” official documentation. link
  30. J. Deng, W. Dong, R. Socher, L. Li, Kai Li, Li Fei-Fei, “ImageNet: A large-scale hierarchical image database,” 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009. source
  31. K. Simonyan, A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” arXiv preprint, 2014. source
  32. K. He, X. Zhang, S. Ren, J. Sun, “Deep Residual Learning for Image Recognition,” arXiv preprint, 2015. source
  33. T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie, “Feature pyramid networks for object detection,” in Proc. IEEE CVPR, 2017. doi
  34. S. Liu, L. Qi, H. Qin, J. Shi and J. Jia, “Path aggregation network for instance segmentation,” in Proc. IEEE/CVF CVPR, pp. 8759–8768, 2018. doi
  35. K. He, X. Zhang, S. Ren, J. Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,” Lecture Notes in Computer Science / Computer Vision – ECCV 2014, 2014. doi · manuscript
  36. C.-Y. Wang, H.-Y. M. Liao, I-H. Yeh, Y.-H. Wu, P.-Y. Chen and J.-W. Hsieh, “CSPNet: A new backbone that can enhance learning capability of CNN,” CVPR Workshops, 2020. arXiv · doi
  37. A. Bochkovskiy and contributors, “Darknet: YOLOv4 configuration and detection layer,” official implementation, accessed 2 November 2026. configuration · assignment and decoding
  38. M. Tan, Q. V. Le, “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” ICML, 2019. source · proceedings
  39. G. Huang, Z. Liu, L. Van Der Maaten, K. Q. Weinberger, “Densely Connected Convolutional Networks,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. doi · manuscript
  40. S. Xie, R. Girshick, P. Dollar, Z. Tu, K. He, “Aggregated Residual Transformations for Deep Neural Networks,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. doi · manuscript
  41. D. Misra, “Mish: A Self Regularized Non-Monotonic Activation Function,” BMVC, 2020. source
  42. P. Ramachandran, B. Zoph, Q. V. Le, “Searching for Activation Functions,” arXiv preprint, 2017. source
  43. J. Yu, Y. Jiang, Z. Wang, Z. Cao, T. Huang, “UnitBox: An Advanced Object Detection Network,” ACM Multimedia, 2016. source
  44. H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid, S. Savarese, “Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regression,” arXiv preprint, 2019. source
  45. Z. Zheng, P. Wang, W. Liu, J. Li, R. Ye and D. Ren, “Distance-IoU loss: Faster and better learning for bounding box regression,” in Proc. AAAI, pp. 12993–13000, 2020. doi
  46. H. Zhang, M. Cisse, Y. N. Dauphin and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” in Proc. ICLR, 2018. arXiv
  47. S. Yun, D. Han, S. Chun, S. J. Oh, Y. Yoo, J. Choe, “CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features,” 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019. doi · manuscript
  48. Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, Ruslan Salakhutdinov, “Dropout: A Simple Way to Prevent Neural Networks from Overfitting,” Journal of Machine Learning Research, 2014. source
  49. G. Ghiasi, T. Lin, Q. V. Le, “DropBlock: A regularization method for convolutional networks,” NeurIPS, 2018. source · proceedings
  50. C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, “Rethinking the Inception Architecture for Computer Vision,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. doi · manuscript
  51. S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” in Proc. ICML, PMLR vol. 37, pp. 448–456, 2015. link
  52. Z. Yao, Y. Cao, S. Zheng, G. Huang, S. Lin, “Cross-Iteration Batch Normalization,” 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. doi · manuscript
  53. S. Woo, J. Park, J. Lee, I. S. Kweon, “CBAM: Convolutional Block Attention Module,” Lecture Notes in Computer Science / Computer Vision – ECCV 2018, 2018. doi · manuscript
  54. I. Loshchilov, F. Hutter, “SGDR: Stochastic Gradient Descent with Warm Restarts,” ICLR, 2017. source
  55. T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár and C. L. Zitnick, “Microsoft COCO: Common objects in context,” in Proc. ECCV, pp. 740–755, 2014. doi
  56. COCO Consortium, “COCO detection evaluation,” official evaluation description. link
  57. ONNX, “Open Neural Network Exchange,” official project page. link
  58. B. Jacob, S. Kligys, B. Chen, M. Zhu, M. Tang, A. Howard, H. Adam and D. Kalenichenko, “Quantization and training of neural networks for efficient integer-arithmetic-only inference,” in Proc. IEEE/CVF CVPR, pp. 2704–2713, 2018. doi
  59. N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov and S. Zagoruyko, “End-to-end object detection with transformers,” in Proc. ECCV, 2020. doi
  60. H. W. Kuhn, “The Hungarian method for the assignment problem,” Naval Research Logistics Quarterly, vol. 2, pp. 83–97, 1955. doi
  61. J. Terven, D.-M. Córdova-Esparza and J.-A. Romero-González, “A comprehensive review of YOLO architectures in computer vision: From YOLOv1 to YOLOv8 and YOLO-NAS,” Machine Learning and Knowledge Extraction, vol. 5, no. 4, pp. 1680–1716, 2023. doi · arXiv
  62. N. Jegham, C. Y. Koh, M. Abdelatti and A. Hendawi, “YOLO evolution: A comprehensive benchmark and architectural review of YOLOv12, YOLO11, and their previous versions,” arXiv:2411.00201, 2024 (revised 2025). arXiv
  63. R. Sapkota and M. Karkee, “Ultralytics YOLO evolution: An overview of YOLO27, YOLO26, YOLO11, YOLOv8, and YOLOv5 object detectors for computer vision and pattern recognition,” arXiv:2510.09653, 2025 (revised 2026). arXiv
  64. T. Shehzadi, K. A. Hashmi, M. Liwicki, D. Stricker and M. Z. Afzal, “Object detection with transformers: A review,” Sensors, vol. 25, no. 19, 2025. doi
  65. Ultralytics, “YOLO27 (coming soon),” official documentation page, accessed 2 November 2026. docs
  66. M. Oquab, T. Darcet, T. Moutakanni, H. Vo, M. Szafraniec et al., “DINOv2: Learning robust visual features without supervision,” arXiv:2304.07193, 2023. arXiv
  67. O. Siméoni, H. V. Vo, M. Seitzer, F. Baldassarre, M. Oquab et al., “DINOv3,” arXiv:2508.10104, 2025. arXiv
  68. X. Ding, X. Zhang, N. Ma, J. Han, G. Ding and J. Sun, “RepVGG: Making VGG-style ConvNets great again,” in Proc. IEEE/CVF CVPR, pp. 13728–13737, 2021. doi
  69. R. Girshick, “Fast R-CNN,” 2015 IEEE International Conference on Computer Vision (ICCV), 2015. doi · manuscript
  70. S. Ren, K. He, R. Girshick, J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” arXiv preprint, 2015. source
  71. M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, A. Zisserman, “The Pascal Visual Object Classes (VOC) Challenge,” International Journal of Computer Vision, 2009. source
  72. A. Shrivastava, A. Gupta, R. Girshick, “Training Region-Based Object Detectors with Online Hard Example Mining,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. doi · manuscript
  73. H. Zhang, Y. Wang, F. Dayoub and N. Sünderhauf, “VarifocalNet: An IoU-aware dense object detector,” in Proc. IEEE/CVF CVPR, pp. 8510–8519, 2021. doi
  74. YOLOX contributors, “YOLOX detection head,” official implementation, accessed 2 November 2026. code
  75. K. Duan, S. Bai, L. Xie, H. Qi, Q. Huang, Q. Tian, “CenterNet: Keypoint Triplets for Object Detection,” arXiv preprint, 2019. source
  76. Z. Ge, S. Liu, Z. Li, O. Yoshie and J. Sun, “OTA: Optimal transport assignment for object detection,” in Proc. IEEE/CVF CVPR, pp. 303–312, 2021. doi
  77. C. Li, L. Li, Y. Geng, et al., “YOLOv6 v3.0: A Full-Scale Reloading,” arXiv preprint, 2023. source
  78. C. Feng, Y. Zhong, Y. Gao, M. R. Scott, W. Huang, “TOOD: Task-Aligned One-Stage Object Detection,” Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021. source
  79. Z. Gevorgyan, “SIoU Loss: More Powerful Learning for Bounding Box Regression,” arXiv preprint, 2022. source
  80. X. Li, W. Wang, L. Wu, S. Chen, X. Hu, J. Li, J. Tang and J. Yang, “Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection,” in Advances in Neural Information Processing Systems 33 (NeurIPS), 2020. proceedings
  81. X. Ding, H. Chen, X. Zhang, K. Huang, J. Han, G. Ding, “Re-parameterizing Your Optimizers rather than Architectures,” ICLR, 2023. source
  82. S. Zhang, C. Chi, Y. Yao, Z. Lei, S. Z. Li, “Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection,” 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020. doi · manuscript
  83. C. Wang, H. M. Liao, I. Yeh, “Designing Network Design Strategies Through Gradient Path Analysis,” arXiv preprint, 2022. source
  84. T. Dao, D. Fu, S. Ermon, A. Rudra, C. Ré, “FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness,” Advances in Neural Information Processing Systems 35, 2022. doi · manuscript
  85. D. P. Kingma, J. Ba, “Adam: A Method for Stochastic Optimization,” ICLR, 2015. source
  86. K. Jordan, “Muon: An optimizer for hidden layers in neural networks,” technical blog, 2024. source
  87. S. Shao, Z. Li, T. Zhang, C. Peng, G. Yu, X. Zhang, J. Li, J. Sun, “Objects365: A Large-Scale, High-Quality Dataset for Object Detection,” Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019. source