Full search page →

YOLOX: anchor-free YOLO with SimOTA

Optional · paper guide 9 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: Z. Ge, S. Liu, F. Wang, Z. Li and J. Sun, “YOLOX: Exceeding YOLO series in 2021,” arXiv, 2021 [15]. Related part of YOLOv4: the YOLOv3 head and box decoding and label assignment.

Challenge

By 2021, detection research had moved to anchor-free detectors, prediction-aware label assignment and end-to-end (NMS-free) training, but YOLOv4 and YOLOv5 [3] still used anchors and hand-crafted assignment rules [15]. The YOLOX report brings those advances into a YOLO. It deliberately starts from YOLOv3 with SPP rather than YOLOv4, because it considers YOLOv4 and YOLOv5 “a little over-optimized for the anchor-based pipeline” [15].

Method

Compare the coupled head with YOLOX’s separate classification and regression paths. The IoU branch belongs to the regression path; this head diagram does not specify the whole detector.
Original paper figure 2 — Compare the coupled head with YOLOX’s separate classification and regression paths. The IoU branch belongs to the regression path; this head diagram does not specify the whole detector.

Figure source: [15], arXiv:2107.08430v2, PDF page 3. Original PDF.

A strong baseline. YOLOv3 with a Darknet-53 backbone and an SPP layer, trained with EMA weight averaging, a cosine learning-rate schedule, an IoU loss with an IoU-aware branch, and BCE for the class and objectness branches; 300 epochs with 5 warm-up epochs, SGD with momentum 0.9, learning rate 0.01×BatchSize/640.01 \times \text{BatchSize}/64, batch 128, inputs drawn from 448 to 832 pixels; only flips, colour jitter and multi-scale inputs for augmentation at this stage [15].

Decoupled head. YOLO heads were coupled: one convolution produced class scores, objectness and box numbers together. YOLOX uses a light decoupled head per FPN level: a 1×11 \times 1 convolution reduces the channels to 256, followed by two parallel branches with two 3×33 \times 3 convolutions each, one for classification and one for regression (with an IoU branch on the regression side) [15]. The motivation is the known conflict between classification and regression features. Figure 4 shows the resulting anatomy, which most later YOLOs share.

Block diagram. A 640 by 640 input feeds a backbone with stages C3, C4 and C5 at strides 8, 16 and 32. A neck turns them into P3, P4 and P5 with orange top-down arrows (FPN) and pink bottom-up arrows (PAN). Each level feeds a decoupled head with a class branch and a box branch, producing 80 by 80, 40 by 40 and 20 by 20 grids, which together give 8,400 candidates. A note says a one-to-many head needs threshold plus NMS, while a one-to-one head uses top-k without NMS.
Figure 4 — The backbone–neck–head design of anchor-free YOLOs after YOLOX, for a 640 × 640 input: an FPN–PAN neck and a decoupled head on strides 8, 16 and 32. Each cell of each grid makes one prediction, giving 8,400 candidates per image (YOLOv4 makes three per cell).

Strong augmentation. Mosaic and MixUp [46], switched off for the last 15 epochs so that training ends on images that look like test images; with this augmentation, ImageNet pre-training was “no more beneficial” and all later models are trained from scratch [15].

Anchor-free. Predictions per location drop from 3 to 1, and the head directly predicts two offsets from the cell’s top-left corner plus the box’s height and width; the centre location of each object is its positive and a scale range assigns objects to FPN levels, as in FCOS [15]. The report gives two reasons against anchors: clustered anchors are domain-specific, and they multiply the number of predictions, which can become a latency bottleneck when moving outputs between devices [15]. Unlike FCOS’s four edge distances, YOLOX predicts a center offset and logarithmic width/height. For grid coordinates (gx,gy)(g_x,g_y) and stride ss, its decoding is cx=(tx+gx)sc_x=(t_x+g_x)s, cy=(ty+gy)sc_y=(t_y+g_y)s, w=etwsw=e^{t_w}s, h=ethsh=e^{t_h}s, where tx,ty,tw,tht_x,t_y,t_w,t_h are raw box outputs. The released head applies no sigmoid to the center offsets; objectness and class probabilities are separate [74]. The snippet uses that parameterization and multiplies objectness by class probability for filtering.

Code · Python · 32 lines
import numpy as np

def make_grid(img_size=640, strides=(8, 16, 32)):
    grids, all_strides = [], []
    for stride in strides:
        size = img_size // stride
        gy, gx = np.mgrid[:size, :size]
        grids.append(np.stack([gx.ravel(), gy.ravel()], axis=1))
        all_strides.append(np.full(size*size, stride))
    return np.concatenate(grids), np.concatenate(all_strides)

def decode(raw, grid, stride, threshold=0.25):
    # raw: center offsets, log(width/stride), log(height/stride), objectness and class logits.
    center = (raw[:, :2] + grid) * stride[:, None]
    size = np.exp(raw[:, 2:4]) * stride[:, None]
    boxes = np.concatenate([center-size/2, center+size/2], axis=1)
    probability = 1/(1+np.exp(-raw[:, 4:]))
    scores = probability[:, :1] * probability[:, 1:]
    classes, best = scores.argmax(1), scores.max(1)
    keep = best > threshold
    return boxes[keep], classes[keep], best[keep]

grid, stride = make_grid()
raw = np.zeros((len(grid), 5+3))
raw[:, 4:] = -6  # background objectness and class logits
index = np.flatnonzero((stride==16) & (grid[:, 0]==18) & (grid[:, 1]==12))[0]
raw[index, :4] = [0., 0.25, np.log(9.), np.log(5.5)]
raw[index, 4] = 4.  # objectness
raw[index, 5+1] = 4.  # class 1
boxes, classes, scores = decode(raw, grid, stride)
print("predictions per 640x640 image:", len(grid))
print("kept:", len(boxes), "| box:", boxes[0], "| class:", classes[0], "| score:", round(float(scores[0]), 3))

Output

predictions per 640x640 image: 8400
kept: 1 | box: [216. 152. 360. 240.] | class: 1 | score: 0.964

The grid location (18,12)(18,12) at stride 16 gives center (288,196)(288,196) after the offset, width 144 and height 88. This is a decoding demonstration, not a trained detector or a reproduction of a benchmark score. YOLOv4’s three anchors per cell would produce three times as many raw box candidates at this resolution.

Multi positives. One positive per object (the centre cell) wastes good predictions, so YOLOX makes the centre 3×33 \times 3 area positive (“centre sampling” in FCOS) [15].

SimOTA. OTA [76] treats label assignment as optimal transport between ground truths and predictions, solved with the Sinkhorn–Knopp algorithm, which adds about 25% training time. SimOTA keeps OTA’s four insights (loss-aware, centre prior, a dynamic number of positives per object, a global view) and approximates the solution: for each pair of ground truth gig_i and prediction pjp_j it computes the cost

cij=Lijcls+λLijreg,c_{ij} = L^{\text{cls}}_{ij} + \lambda L^{\text{reg}}_{ij},

where LijclsL^{\text{cls}}_{ij} and LijregL^{\text{reg}}_{ij} are the classification and regression losses between them and λ\lambda a balancing coefficient, and selects for each ground truth the kk lowest-cost predictions within a fixed centre region as positives, with kk chosen per object by OTA’s dynamic-kk estimation [15].

Optional end-to-end variant. Two extra convolutional layers, one-to-one label assignment and a stop-gradient make the detector NMS-free, but slightly less accurate and slower, so it is left out of the final models [15].

Results

All latencies are measured with FP16 precision and batch 1 on a single Tesla V100, and the roadmap table excludes post-processing [15]. On COCO test-dev, YOLOX-L reaches 50.0% AP at 69.0 FPS and YOLOX-DarkNet53 47.4% AP at 90.1 FPS; in the same table, YOLOv4-CSP is listed at 47.5% AP and 73.0 FPS and YOLOv5-L at 48.2% AP and 73.0 FPS [15]. At the small end, YOLOX-Nano with 0.91 M parameters and 1.08 GFLOPs reaches 25.3% AP on COCO val [15]. YOLOX won the Streaming Perception Challenge at the CVPR 2021 autonomous-driving workshop with a single YOLOX-L model [15].

Ablations

Table 2, the roadmap of YOLOX-DarkNet53 on COCO val at 640 × 640 [15]:

StepAPLatency (V100, FP16, batch 1, no post-processing)
YOLOv3 baseline38.510.5 ms
+ decoupled head39.611.6 ms
+ strong augmentation42.011.6 ms
+ anchor-free42.911.1 ms
+ multi positives45.011.1 ms
+ SimOTA47.311.1 ms
+ NMS-free (optional)46.513.5 ms

Table 1 shows that the decoupled head matters even more for the end-to-end version: with a coupled head, going end to end costs 4.2 AP (38.5 → 34.3), with a decoupled head only 0.8 AP (39.6 → 38.8) [15]. The decoupled head also converges much faster (the training curves of Figure 3). Table 5 shows that strong augmentation should scale with model size: removing MixUp and weakening Mosaic’s scale range improves YOLOX-Nano from 24.0% to 25.3% AP, while scale-jittered MixUp improves YOLOX-L from 48.6% to 49.5% [15].

Conclusion

YOLOX shows that the anchor-based head YOLOv4 kept is not necessary for a fast YOLO: an anchor-free decoupled head with centre sampling and SimOTA, trained with strong augmentation from scratch, adds about 9 AP to a YOLOv3 baseline at nearly the same latency.

Problems

It is an arXiv report; its NMS-free variant loses accuracy and speed, so YOLOX still depends on NMS; SimOTA’s dynamic kk adds training cost and, according to the YOLOv6 report, can slow training and make it unstable [16]. Its latencies exclude post-processing. YOLOv6 replaced SimOTA with task-aligned assignment; YOLOv10 made the NMS-free head competitive.

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