pith. sign in

arxiv: 2605.02284 · v1 · submitted 2026-05-04 · 💻 cs.CV

Beyond Known Objects: A Novel Framework for Open-Set Object Detection using Negative-Aware Norm

Yuchen Zhang , Yao Lu , Johannes Betz This is my paper

Pith reviewed 2026-05-09 15:49 UTC · model grok-4.3

classification 💻 cs.CV
keywords open-set object detectionnegative-aware normunknown object detectionobjectness estimationCOCO-Open datasetautonomous drivingpre-trained detectors
0
0 comments X

The pith

Standard detectors already hold cues for unknown objects via a hidden-layer metric that needs almost no extra training.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that detectors trained only on known object categories still develop internal signals useful for spotting any object. It introduces NAN-SPOT, a framework that extracts a Negative-Aware Norm from a hidden layer to score objectness and requires only minutes of training on a few hundred images. This avoids the heavy retraining that most open-set methods demand. Tests on an expanded COCO-Open dataset with 1853 unknown annotations show better unknown-object detection than retraining-heavy baselines while keeping known-object performance intact. The result matters for autonomous driving systems that must handle novel obstacles in changing scenes.

Core claim

NAN-SPOT shows that computing the Negative-Aware Norm from a hidden layer of a frozen off-the-shelf detector estimates objectness well enough to surpass methods that retrain the detector extensively for open-set detection, while using far less data and time.

What carries the argument

Negative-Aware Norm (NAN), a metric from a hidden layer that gauges objectness by incorporating information from negative samples.

If this is right

  • Unknown-object detection exceeds the results of methods that retrain the full detector.
  • Accuracy on known objects remains unchanged.
  • Training takes minutes and uses only hundreds of images.
  • COCO-Open supplies 1853 unknown annotations for more complete evaluation than prior datasets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Many existing deployed detectors could gain open-set ability through a lightweight add-on rather than full replacement.
  • Objectness may be a general latent property rather than one tied only to the specific training categories.
  • The same norm-based approach could be tested on detectors for robotics or surveillance to check broader applicability.

Load-bearing premise

Training on many known categories has already imprinted useful objectness cues into the hidden layers of standard detectors.

What would settle it

Applying the Negative-Aware Norm to a detector trained on only a narrow set of categories and finding it performs no better than chance at separating unknowns on the expanded COCO-Open set would disprove the central premise.

Figures

Figures reproduced from arXiv: 2605.02284 by Johannes Betz, Yao Lu, Yuchen Zhang.

Figure 1
Figure 1. Figure 1: Top: NAN-SPOT extends D-DETR with an objectness module estimating objectness scores for each bounding box, supported by the NAN metric derived from the detector’s final hidden layer. Bottom: Predicted unknown/ known objects and training time of CAT [7], PROB [8] and our model. To compensate for these perceptual uncertainties, deployed systems often rely on sensor redundancy and conservative fallback strate… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed NAN-SPOT for open-set object detection. (Bottom) Based on D-DETR, which extracts multi-scale feature maps from a CNN backbone and encodes them using a transformer encoder with deformable attention to focus on sparse, informative regions. A set of learnable query embeddings is then passed to the decoder, where they are iteratively refined via cross-attention with the encoded feature… view at source ↗
Figure 3
Figure 3. Figure 3: A proof-of-concept for D-DETR metrics on COCO-Open. (left) Distribution of metrics across known /unknown objects and background regions. KDE is applied to estimate the probability density. (a) Confidence shows substantial overlap between unknowns and background. (b) NAN metric assign higher scores to unknown but introduces confusion for background. (c) Objectness shows clear separation, with known and unkn… view at source ↗
Figure 4
Figure 4. Figure 4: Top 10 detections by NAN. While successfully assigning high scores to salient objects (e.g., both zebras), NAN also consistently assigns high scores to tiny regions and bounding boxes along the edges of the image. required for full detector retraining. In principle, g(·) may be realized using any classification method. In our experiments, we evaluate two lightweight estimators: a random forest [34] and a m… view at source ↗
Figure 5
Figure 5. Figure 5: Examples from COCO-Mixed illustrating distinct limitations (a–c), and an improved annotation from COCO-Open in (d). (a) Images with composite items (e.g., soup) present ambiguity regarding whether smaller components (e.g., vegetables, meat) should be annotated as distinct objects, making them unsuitable for evaluation. (b) Visual evidence is insufficient to confirm whether the labeled objects are indeed a … view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results on CODA. Comparison of OW-DETR [17], PROB [8], HYP-OW [18] and NAN-SPOT on detecting known and unknown objects. Same number of top-k predictions are shown per image for fair comparison. 2) Qualitative Analysis view at source ↗
Figure 7
Figure 7. Figure 7 view at source ↗
read the original abstract

Open-Set Object Detection (OSOD) is crucial for autonomous driving, where perception systems must recognize and localize both known and previously unseen objects in complex, dynamic environments. While recent approaches deliver promising results, they often require retraining the detector extensively to learn objectness, which describes the likelihood that a bounding box tightly encloses a valid object, regardless of whether its category was learned during training. Deviating from existing work, we hypothesize that standard off-the-shelf detectors may already contain helpful cues for objectness, owing to their training on numerous and diverse known categories. Building on this idea, we propose NAN-SPOT, a training-light framework that does not require to retrain the base object detector and estimates objectness by leveraging a hidden layer metric called Negative-Aware Norm (NAN), requiring only minutes of training on just hundreds of images. To support comprehensive evaluation, we introduce COCO-Open, an expanded version of the existing COCO-Mixed dataset, increasing unknown object annotations from 433 to 1853, making it the most exhaustively labeled dataset for OSOD to the best of our knowledge. Experimental results demonstrate that NAN-SPOT achieves even better performance on unknown object detection than methods requiring heavy training, without compromising performance on known objects. This efficiency and robustness make NAN-SPOT a promising step towards open-world perception in autonomous driving.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes NAN-SPOT, a training-light open-set object detection framework that extracts a Negative-Aware Norm (NAN) metric from a hidden layer of an off-the-shelf detector to estimate objectness without retraining the base model. It introduces the COCO-Open dataset (expanding unknown annotations from 433 to 1853) and claims that NAN-SPOT outperforms heavily retrained OSOD baselines on unknown objects while preserving known-object performance, with only minutes of calibration on a few hundred images.

Significance. If the empirical claims hold, the work would be significant for autonomous driving and open-world perception by demonstrating that category-agnostic objectness cues may already exist in standard detectors, enabling efficient OSOD without the computational cost of full retraining. The expanded COCO-Open dataset is a clear positive contribution for future benchmarking.

major comments (3)
  1. [Abstract and §4] Abstract and §4 (Experiments): The central claim that NAN-SPOT 'achieves even better performance on unknown object detection than methods requiring heavy training' is unsupported by any quantitative results, baselines, error bars, or evaluation protocol details in the provided text, preventing verification against the skeptic's concern that gains may be dataset artifacts.
  2. [§3] §3 (Method): The Negative-Aware Norm is introduced as encoding a category-agnostic objectness signal, but no distribution plots, statistical separation tests, or ablations on hidden-layer choice are described to confirm it distinguishes unknowns from background clutter rather than category-specific features.
  3. [§4] §4 (Experiments): No ablation on the light calibration set (hundreds of images) is reported to rule out overfitting to the particular unknown instances in COCO-Open, which is load-bearing for the claim that NAN generalizes beyond the calibration data.
minor comments (2)
  1. [Abstract] The abstract refers to 'COCO-Mixed dataset' without a citation; a reference to the original source should be added.
  2. [§3] Notation for NAN is introduced without an explicit equation; adding a formal definition (e.g., Eq. (X)) would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed review of our manuscript. We address each of the major comments point by point below, and we will incorporate revisions to strengthen the presentation and support for our claims.

read point-by-point responses

  1. Referee: [Abstract and §4] The central claim that NAN-SPOT 'achieves even better performance on unknown object detection than methods requiring heavy training' is unsupported by any quantitative results, baselines, error bars, or evaluation protocol details in the provided text, preventing verification against the skeptic's concern that gains may be dataset artifacts.

    Authors: We agree that additional quantitative details are necessary to fully substantiate the performance claims. The manuscript does report experimental results in §4, but to address this concern directly, we will revise the abstract and §4 to include explicit comparison tables with baselines, error bars from repeated experiments, and a detailed evaluation protocol description. This will allow readers to verify the results and confirm that improvements on unknown objects are robust. revision: yes

  2. Referee: [§3] The Negative-Aware Norm is introduced as encoding a category-agnostic objectness signal, but no distribution plots, statistical separation tests, or ablations on hidden-layer choice are described to confirm it distinguishes unknowns from background clutter rather than category-specific features.

    Authors: To provide evidence that NAN represents a category-agnostic objectness signal rather than category-specific features, we will augment §3 with distribution plots of NAN scores across known objects, unknown objects, and background clutter. Additionally, we will include statistical separation tests (e.g., t-tests or KS-tests) and ablations over different hidden layers to demonstrate the generality of the chosen metric. revision: yes

  3. Referee: [§4] No ablation on the light calibration set (hundreds of images) is reported to rule out overfitting to the particular unknown instances in COCO-Open, which is load-bearing for the claim that NAN generalizes beyond the calibration data.

    Authors: We recognize the value of such an ablation for validating generalization. In the revised manuscript, we will add experiments in §4 that ablate the calibration set size and composition, using varying numbers of images and different selections of unknown instances, to show that performance on unseen unknowns remains consistent and does not rely on overfitting to the calibration data. revision: yes

Circularity Check

0 steps flagged

No circularity: proposal rests on empirical hypothesis and external evaluation, not self-referential definitions or fitted predictions

full rationale

The paper advances a hypothesis that off-the-shelf detectors already encode objectness cues in hidden layers, then defines NAN as a simple norm-based metric extracted from those layers and evaluates it on an expanded dataset. No equations, derivations, or parameter-fitting steps are described that would reduce the claimed performance gains to the inputs by construction. The central claim is supported by comparative experiments rather than any self-citation chain or ansatz smuggled from prior author work. This is a standard empirical framework paper with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract-only review supplies insufficient technical detail to enumerate free parameters, axioms, or invented entities beyond the high-level introduction of the NAN metric and COCO-Open dataset.

invented entities (1)
  • Negative-Aware Norm (NAN) no independent evidence
    purpose: Metric extracted from a hidden layer of a pre-trained detector to estimate objectness without retraining
    Presented as the core novel component enabling the training-light framework

pith-pipeline@v0.9.0 · 5546 in / 1158 out tokens · 45670 ms · 2026-05-09T15:49:53.308940+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [1]

    Yolov8: A novel object detection algo- rithm with enhanced performance and robustness,

    R. Varghese and M. Sambath, “Yolov8: A novel object detection algo- rithm with enhanced performance and robustness,” in2024 International conference on advances in data engineering and intelligent computing systems. IEEE, 2024, pp. 1–6

    work page 2024

  2. [2]

    Center-based 3d object detection and tracking,

    T. Yin, X. Zhou, and P. Krahenbuhl, “Center-based 3d object detection and tracking,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 11 784–11 793

    work page 2021

  3. [3]

    Bevformer: Learning bird’s-eye-view representation from lidar-camera via spatiotemporal transformers,

    Z. Li, W. Wang, H. Li, E. Xie, C. Sima, T. Lu, Q. Yu, and J. Dai, “Bevformer: Learning bird’s-eye-view representation from lidar-camera via spatiotemporal transformers,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 3, pp. 2020–2036, 2025

    work page 2020

  4. [4]

    Anomaly detection in autonomous driving: A survey,

    D. Bogdoll, M. Nitsche, and J. M. Z ¨ollner, “Anomaly detection in autonomous driving: A survey,” inProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, 2022, pp. 4488–4499

    work page 2022

  5. [5]

    Foundation models in autonomous driving: A survey on scenario generation and scenario analysis,

    Y . Gao, M. Piccinini, Y . Zhang, D. Wang, K. Moller, R. Brusnicki, B. Zarrouki, A. Gambi, J. F. Totz, K. Stormset al., “Foundation models in autonomous driving: A survey on scenario generation and scenario analysis,”arXiv preprint arXiv:2506.11526, 2025

    work page arXiv 2025

  6. [6]

    On perceptual uncertainty in autonomous driving under consideration of contextual awareness,

    A. Saad, N. Bangalore, I. Kurzidem, and P. Schleiss, “On perceptual uncertainty in autonomous driving under consideration of contextual awareness,” in2022 6th International Conference on System Reliability and Safety (ICSRS). IEEE, 2022, pp. 387–393

    work page 2022

  7. [7]

    Cat: Localization and identification cascade detection transformer for open- world object detection,

    S. Ma, Y . Wang, Y . Wei, J. Fan, T. H. Li, H. Liu, and F. Lv, “Cat: Localization and identification cascade detection transformer for open- world object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 19 681–19 690

    work page 2023

  8. [8]

    Prob: Probabilistic objectness for open world object detection,

    O. Zohar, K.-C. Wang, and S. Yeung, “Prob: Probabilistic objectness for open world object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 11 444–11 453

    work page 2023

  9. [9]

    Unknown-aware object detection: Learning what you don’t know from videos in the wild,

    X. Du, X. Wang, G. Gozum, and Y . Li, “Unknown-aware object detection: Learning what you don’t know from videos in the wild,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 13 678–13 688

    work page 2022

  10. [10]

    Detecting the unknown in object detection,

    D. Fontanel, M. Tarantino, F. Cermelli, and B. Caputo, “Detecting the unknown in object detection,”CoRR, vol. abs/2208.11641, 2022

    work page arXiv 2022

  11. [11]

    Uadet: A remarkably simple yet effective uncertainty-aware open-set object detection framework,

    S. Cheng, Y . Liu, and K. Han, “Uadet: A remarkably simple yet effective uncertainty-aware open-set object detection framework,”CoRR, vol. abs/2412.09229, 2024

    work page arXiv 2024

  12. [12]

    The overlooked elephant of object detection: Open set,

    A. Dhamija, M. Gunther, J. Ventura, and T. Boult, “The overlooked elephant of object detection: Open set,” inProceedings of the IEEE/CVF winter conference on applications of computer vision, 2020, pp. 1021– 1030

    work page 2020

  13. [13]

    Unknown sniffer for object detection: Don’t turn a blind eye to unknown objects,

    W. Liang, F. Xue, Y . Liu, G. Zhong, and A. Ming, “Unknown sniffer for object detection: Don’t turn a blind eye to unknown objects,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 3230–3239

    work page 2023

  14. [14]

    Novel scenes & classes: Towards adaptive open-set object detection,

    W. Li, X. Guo, and Y . Yuan, “Novel scenes & classes: Towards adaptive open-set object detection,” inIEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, October 1-6, 2023. IEEE, 2023, pp. 15 734–15 744

    work page 2023

  15. [15]

    Opengan: Open-set recognition via open data generation,

    S. Kong and D. Ramanan, “Opengan: Open-set recognition via open data generation,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 813–822

    work page 2021

  16. [16]

    To- wards open world object detection,

    K. J. Joseph, S. H. Khan, F. S. Khan, and V . N. Balasubramanian, “To- wards open world object detection,” inIEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021. Computer Vision Foundation / IEEE, 2021, pp. 5830–5840

    work page 2021

  17. [17]

    OW-DETR: open-world detection transformer,

    A. Gupta, S. Narayan, K. J. Joseph, S. Khan, F. S. Khan, and M. Shah, “OW-DETR: open-world detection transformer,” inIEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. IEEE, 2022, pp. 9225–9234

    work page 2022

  18. [18]

    Hyp- ow: Exploiting hierarchical structure learning with hyperbolic distance enhances open world object detection,

    T. Doan, X. Li, S. Behpour, W. He, L. Gou, and L. Ren, “Hyp- ow: Exploiting hierarchical structure learning with hyperbolic distance enhances open world object detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 2, 2024, pp. 1555– 1563

    work page 2024

  19. [19]

    Understanding the feature norm for out-of-distribution detection,

    J. Park, J. C. L. Chai, J. Yoon, and A. B. J. Teoh, “Understanding the feature norm for out-of-distribution detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 1557–1567

    work page 2023

  20. [20]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision. Springer, 2020, pp. 213– 229

    work page 2020

  21. [21]

    Deformable detr: De- formable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: De- formable transformers for end-to-end object detection,” inInternational Conference on Learning Representations, 2021

    work page 2021

  22. [22]

    Idpd: Improved deformable-detr for crowd pedestrian detection,

    W. Han, N. He, X. Wang, F. Sun, and S. Liu, “Idpd: Improved deformable-detr for crowd pedestrian detection,”Signal, Image and Video Processing, vol. 18, no. 3, pp. 2243–2253, 2024

    work page 2024

  23. [23]

    Airport uav and birds detection based on deformable detr,

    L. Shanliang, L. Yunlong, Q. Jingyi, and W. Renbiao, “Airport uav and birds detection based on deformable detr,” inJournal of Physics: Conference Series, vol. 2253, no. 1. IOP Publishing, 2022, p. 012024

    work page 2022

  24. [24]

    Accurate leukocyte detection based on deformable-detr and multi-level feature fusion for aiding diagnosis of blood diseases,

    Y . Chen, C. Zhang, B. Chen, Y . Huang, Y . Sun, C. Wang, X. Fu, Y . Dai, F. Qin, Y . Penget al., “Accurate leukocyte detection based on deformable-detr and multi-level feature fusion for aiding diagnosis of blood diseases,”Computers in biology and medicine, vol. 170, p. 107917, 2024

    work page 2024

  25. [25]

    Detreg: Unsupervised pretraining with region priors for object detection,

    A. Bar, X. Wang, V . Kantorov, C. J. Reed, R. Herzig, G. Chechik, A. Rohrbach, T. Darrell, and A. Globerson, “Detreg: Unsupervised pretraining with region priors for object detection,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 14 605–14 615

    work page 2022

  26. [26]

    Dino: Detr with improved denoising anchor boxes for end-to-end object detection,

    H. Zhang, F. Li, S. Liu, L. Zhang, H. Su, J. Zhu, L. Ni, and H.-Y . Shum, “Dino: Detr with improved denoising anchor boxes for end-to-end object detection,” inThe Eleventh International Conference on Learning Representations

    work page

  27. [27]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137–1149, 2016

    work page 2016

  28. [28]

    Exploring orthogonality in open world object detection,

    Z. Sun, J. Li, and Y . Mu, “Exploring orthogonality in open world object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 17 302–17 312

    work page 2024

  29. [29]

    Reducing network agnosto- phobia,

    A. R. Dhamija, M. G ¨unther, and T. Boult, “Reducing network agnosto- phobia,”Advances in Neural Information Processing Systems, vol. 31, 2018

    work page 2018

  30. [30]

    Out-of-distribution detection for reliable face recognition,

    C. Yu, X. Zhu, Z. Lei, and S. Z. Li, “Out-of-distribution detection for reliable face recognition,”IEEE Signal Processing Letters, vol. 27, pp. 710–714, 2020

    work page 2020

  31. [31]

    Norm-aware embedding for efficient person search,

    D. Chen, S. Zhang, J. Yang, and B. Schiele, “Norm-aware embedding for efficient person search,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 12 615–12 624

    work page 2020

  32. [32]

    Magface: A universal repre- sentation for face recognition and quality assessment,

    Q. Meng, S. Zhao, Z. Huang, and F. Zhou, “Magface: A universal repre- sentation for face recognition and quality assessment,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 14 225–14 234

    work page 2021

  33. [33]

    Unknown-aware hierarchical object detection in the context of automated driving,

    J. Zhou, N. Wandelburg, and J. Beyerer, “Unknown-aware hierarchical object detection in the context of automated driving,” in2023 IEEE 26th International Conference on Intelligent Transportation Systems. IEEE, 2023, pp. 2501–2508

    work page 2023

  34. [34]

    Random forests,

    L. Breiman, “Random forests,”Machine learning, vol. 45, pp. 5–32, 2001

    work page 2001

  35. [35]

    Learning repre- sentations by back-propagating errors,

    D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning repre- sentations by back-propagating errors,”nature, vol. 323, no. 6088, pp. 533–536, 1986

    work page 1986

  36. [36]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” inEuropean conference on computer vision. Springer, 2014, pp. 740–755

    work page 2014

  37. [37]

    Lvis: A dataset for large vocabulary instance segmentation,

    A. Gupta, P. Dollar, and R. Girshick, “Lvis: A dataset for large vocabulary instance segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 5356– 5364

    work page 2019

  38. [38]

    Expanding low-density latent regions for open-set object detection,

    J. Han, Y . Ren, J. Ding, X. Pan, K. Yan, and G. Xia, “Expanding low-density latent regions for open-set object detection,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. IEEE, 2022, pp. 9581–9590

    work page 2022

  39. [39]

    Dropout sampling for robust object detection in open-set conditions,

    D. Miller, L. Nicholson, F. Dayoub, and N. S ¨underhauf, “Dropout sampling for robust object detection in open-set conditions,” in2018 IEEE International Conference on Robotics and Automation. IEEE, 2018, pp. 3243–3249

    work page 2018

  40. [40]

    Coda: A real-world road corner case dataset for object detection in autonomous driving,

    K. Li, K. Chen, H. Wang, L. Hong, C. Ye, J. Han, Y . Chen, W. Zhang, C. Xu, D.-Y . Yeunget al., “Coda: A real-world road corner case dataset for object detection in autonomous driving,”arXiv preprint arXiv:2203.07724, 2022

    work page arXiv 2022