Pith. sign in

REVIEW 3 major objections 4 minor 36 references

Assured Autonomy with Neuro-Symbolic Perception

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Scene graphs expose a LiDAR attack that fools camera–LiDAR fusion.

desk verdict A plausible integration of scene-graph consistency for attack detection, but the feasibility evidence is anecdotal and the abstract overclaims; the idea deserves a serious referee with major revision. read the letter →

arxiv 2505.21322 v1 pith:FLTLQV3X submitted 2025-05-27 cs.AI

classification cs.AI
keywords neuro-symbolicperceptionscenegraphgenerationsensorfusionsecurityfrustumattackautonomousdrivingLiDARcyber-physicalsystems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that deep-learning perception for autonomous vehicles is fundamentally insecure because DNNs are equivariant pattern-matchers: an attacker who shifts a 3D object in LiDAR data while keeping its projection consistent with the camera can move it tens of meters without tripping standard fusion checks. To close this gap, the authors propose a neuro-symbolic perception paradigm (NeuSPaPer) in which object detection is paired with scene graph generation, so each sensor yields not just boxes but structured relations such as "van in front of truck." They report feasibility studies on the CARLA simulator and nuScenes dataset showing that a camera-derived scene graph and a LiDAR-derived scene graph agree on benign scenes, and that the same cross-sensor graph comparison flags the translated object when the LiDAR graph is attacked. If correct, this is the first single-platform method that detects frustum-style attacks exploiting asymmetric sensor resolutions.

What carries the argument

The load-bearing mechanism is cross-sensor scene graph consistency evaluation. A scene graph is a graph whose nodes are detected objects and whose edges are relational predicates such as front of, left of, near, and occluding; scene graph generation (SGG) is the joint task of detecting objects and inferring these relations. For the camera, graphs are produced by a vision-language foundation model prompted in natural language; for LiDAR, graphs are produced by hand-coded geometric functions over 3D box detections. Per-sensor integrity checks each graph against physics- and commonsense-based knowledge, and a cross-sensor integrity stage matches nodes between graphs and flags edges that disagree. This graph-level comparison is what lifts 2D image data into relational 3D space and exposes semantic changes that box-level IoU checks cannot see.

What would settle it

Run the pipeline on a large set of benign CARLA and nuScenes scenes and count cross-sensor inconsistencies; if the camera foundation-model graph and the LiDAR rule-based graph disagree frequently on un-attacked scenes, high false-alarm rates would mean the proposed comparison cannot distinguish attack-induced semantic changes from ordinary model noise.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that representing sensor output as relational scene graphs turns a previously stealthy attack into a detectable semantic inconsistency. A frustum attack translates an object's 3D box (for example, a van) so the projected box keeps high IoU with the unattacked camera detection; standard single-frame fusion checks therefore pass. But the attack moves the van in 3D space, changing its geometric relations to other objects. The LiDAR rule-based graph then shows, for example, that the van-truck relation flips, while the camera graph, built by a foundation model prompted to "build a scene graph from this image," preserves the original spatial relation. Comparing the two graphs exposes the mismatch, so the authors claim the neuro-symbolic SGG pipeline offers the first method securing perception against attacks that exploit asymmetric sensor resolutions.

Load-bearing premise

The camera graph must be a trustworthy reference: the foundation model's scene graph has to be accurate, stable, and comparable to the LiDAR geometric graph, but the paper does not measure how often the two graphs agree on benign scenes.

Editorial extensions

If this is right

  • Frustum-type translation attacks on LiDAR become detectable even when the attacked boxes keep high overlap with the unattacked camera image.
  • The same graph-based integrity reasoning can flag semantic-structure attacks beyond translations, such as false positives or false negatives that alter relational context.
  • Graph-informed fusion can use per-node and per-edge consistency classifications to weight or reject unreliable sensor updates during state estimation.
  • If specialized transformer-based SGG models replace foundation models, the approach can move toward real-time deployment with bounded output spaces.

Reading between the lines

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

  • Editorial inference: the same cross-sensor graph comparison could extend to other modalities such as radar or thermal cameras, since any sensor that can produce relational structure could be checked against the camera-derived graph.
  • Editorial inference: the approach could also detect spoofed objects that are internally consistent with one sensor but break relational constraints across sensors, not only translated existing objects.
  • Editorial inference: a quantitative benign-scene false-alarm study is the natural next experiment, because the practical detection threshold of the graph comparison is undefined without measuring how often the two graph types naturally disagree.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes NeuSPaPer, a neuro-symbolic perception framework that augments object detection with scene graph generation (SGG) and uses per-sensor and cross-sensor integrity checks to detect semantic-structure attacks, in particular the frustum attack on LiDAR-camera fusion. The authors derive an optimal frustum attack (Eq. 1), propose rule-based SGG for LiDAR and foundation-model SGG for camera images, and present a qualitative case study from CARLA and two from nuScenes. They claim that this is the first single-platform method to detect such attacks, and they discuss challenges and future work toward full-stack neuro-symbolic perception.

Significance. The idea of using scene graphs as a relational integrity layer for multi-sensor perception is attractive, and the attack formulation in Eq. (1) is consistent with prior work. If the central claim were established with quantitative evidence, this would be a meaningful step toward defending fusion pipelines against semantic-structure attacks. However, the evaluated evidence is only qualitative, and the unmeasured camera branch makes the core comparison untested. The paper is best viewed as a research vision with a feasibility illustration, not as a validated security mechanism.

major comments (3)
  1. [Section 5.3, Fig. 6] The central claim that the neuro-symbolic SGG pipeline 'offers the first method that secures perception against attacks exploiting asymmetric sensor resolutions' is not supported by the reported feasibility study, which contains no detection rate, no false-positive analysis, no benign-scene baseline, and no error bars. The demonstration is a single CARLA frame plus two nuScenes frames in Appendix D, and the inconsistency highlighted in Fig. 6(d) is not compared against the distribution of graph mismatches that would occur in unattacked scenes. Without such a comparison, the observed mismatch cannot be attributed to the frustum attack rather than to ordinary sensor noise, detection error, or vocabulary differences between the two graph types.
  2. [Section 5.1, Figure 4 caption, Appendix C.3] The cross-sensor consistency check requires that the foundation-model scene graph from the camera and the rule-based scene graph from LiDAR are commensurable, but the paper never provides an ontology mapping between the foundation model's relation vocabulary and the geometric rules (front_of, left_of, occluding, following, far_from, close_to, next_to). The Figure 4 caption states that 'node and edge types differ from rule-based approach,' and no benign-scene study demonstrates that the two graphs normally agree. If the foundation model says 'beside' where the rule-based graph says 'front_of,' or fails to detect the distant truck in Fig. 6(a), the cross-sensor evaluator will flag benign data as attack, so the load-bearing assertion of Section 5.3 is unverified.
  3. [Section 7] The conclusion concedes that 'uncertain or incomplete graphs due to noisy data can yield inaccurate conclusions,' which directly undermines the feasibility demonstration, because the paper does not quantify graph noise on benign scenes or show that the cross-sensor integrity function separates noise from attack. This concession should either be backed by a false-alarm analysis on benign frames or the Section 5.3 claim should be downgraded from 'first method that secures perception' to a preliminary illustration.
minor comments (4)
  1. [Appendix D] The text says 'Additional analyses and case studies are provided in Appendix D,' but Appendix D contains only two additional qualitative case studies with no numeric analysis; please align the wording with the content.
  2. [Section 5.1] The foundation model is never named; to make the study reproducible, specify the model, version, prompt details, and any sampling parameters.
  3. [Appendix C.3] The list of relations includes 'far from' and 'close to' with complement 'self'; the notion of a 'reduced graph' and how complements are handled for these self-complement relations should be clarified.
  4. [Section 3.1] The statement that 'no amount of sub-sampling or ensembling can mitigate translation attacks' is too categorical; the paper should precisely state the class of defenses it rules out and cite the relevant impossibility or possibility results.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the cross-sensor graph comparison uses an independently pre-trained foundation model plus hand-coded geometric rules, so the attack-induced mismatch is the intended detection mechanism rather than an input-output tautology; the remaining concerns are validation gaps, not circular reasoning.

full rationale

The paper's central demonstration (Secs. 5.1-5.3, Figs. 6-8) does not fit any parameter to the attack and does not present a fitted quantity as a prediction. The LiDAR branch is a deterministic rule-based scene graph over 3D boxes (Sec. 5.1 and Appendix C.3); the camera branch is an externally pre-trained vision-language foundation model prompted with 'Build a scene graph from this image' (Sec. 5.1), which the authors neither train nor tune. The frustum attack is re-derived in Eq. (1) of Appendix B.4 rather than being assumed from a self-citation. The attack's 3D box translation necessarily alters the hand-coded geometric relations; flagging that change through cross-sensor edge matching (Sec. 4.2.3) is the intended operation of the proposed detector, not a circular prediction. The paper does not claim the graph mismatch is a learned or fitted output; it is a feasibility observation. The genuine weaknesses -- no benign-scene false-alarm baseline, unmeasured SGG accuracy of the foundation model, and the acknowledged 'node and edge types differ' between graph modalities (Fig. 4 caption; Sec. 7: 'uncertain or incomplete graphs due to noisy data can yield inaccurate conclusions') -- are empirical validation gaps, not circular reasoning. Self-citations to Hallyburton et al. (2022, 2023a) supply the attack and stealth framework, but the formal optimization is restated in the paper, so they are not load-bearing circular evidence.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on premises the reader must grant. First, point-cloud detectors are translation equivariant in a way that defeats certified robustness (Section 3.1, citing Bronstein et al. 2021). Second, the chosen frustum attack from the authors' own prior work is an accurate model of a realistic adversary. Third, and most load-bearing, the foundation model's image-derived scene graph is a reliable, commensurable reference for cross-sensor comparison; this is not validated. Fourth, the hand-coded geometric rules, whose thresholds are unreported, produce the correct LiDAR graph. The relation thresholds and the cross-sensor consistency tolerance are free parameters with undisclosed values, so the detection boundary is undefined in the text. The external foundation model provides the only independent anchor, which tempers but does not remove the self-referential structure of the demonstration.

free parameters (3)
  • Spatial-relation thresholds for rule-based LiDAR graphs = not reported (deferred to unreleased code)
    Appendix C.3 defines relations such as front of, left of, near and far but reports no distance or angle cutoffs; these thresholds determine the LiDAR graph and therefore which mismatches the integrity check flags.
  • Cross-sensor graph consistency tolerance = not reported
    Section 4.2.3 says unmatched edges are 'evaluated to determine if the lack of agreement is a product of noisy sensor data or an attack' but gives no criterion or threshold, leaving detection and false-positive behavior undefined.
  • IoU stealth threshold zeta_min for the frustum attack = about 0.9 in Figure 1 (from prior work)
    Equation 1 uses a minimum IoU to keep the attacked 3D box consistent with 2D detections; this is an attacker-side parameter inherited from Hallyburton et al. (2022), but it controls how far objects can move in the demonstrated scenes.
assumptions (5)
  • domain assumption Point-cloud perception networks are translation equivariant, so features of a translated object are retained at the new location.
    Section 3.1, citing Bronstein et al. (2021); underpins the claim that randomized smoothing and ensembling cannot stop translation attacks. Plausible and cited, but stated informally (invariant and equivariant are conflated) and not formally derived in this paper.
  • domain assumption Certified robustness guarantees are limited to bounded Lp perturbations and therefore do not cover semantic relocation attacks.
    Section 3.1; used to justify that reactive hardening cannot defend the considered attack class. Consistent with the cited certification literature, but the Lp magnitude of the translation attack is not analyzed here.
  • domain assumption A vision-language foundation model prompted with 'Build a scene graph from this image' yields spatial relationships that are reliable and commensurable with geometric relations computed from 3D boxes.
    Sections 5.1 and 4.2.3; the camera graph is the reference side of the cross-sensor integrity check, yet the model is unnamed, its relation vocabulary differs (Figure 4 caption), and no accuracy validation is given. This is the most load-bearing unvalidated premise.
  • standard math The assignment-problem and IoU-projection formulations (Appendices A and B) correctly model camera-LiDAR fusion and the frustum attack.
    Standard optimization formulation inherited from prior work (Hallyburton et al. 2022); used to construct the demonstrated attack.
  • ad hoc to paper In the case studies, benign camera and LiDAR graphs agree, so a graph mismatch is an attack signal rather than sensor noise.
    Sections 5.3 and Appendix D compare graphs by inspection; there is no benign-scene false-positive study, no noise model, and no adverse-weather or occlusion cases. The 'inconsistency implies attack' interpretation is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Assured Autonomy with Neuro-Symbolic Perception." pith.science (2026). https://pith.science/paper/FLTLQV3X

@misc{pith2026250521322,
  author       = {Pith},
  title        = {Pith review of: Assured Autonomy with Neuro-Symbolic Perception},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FLTLQV3X}},
  note         = {Machine review of arXiv:2505.21322}
}
read the original abstract

Many state-of-the-art AI models deployed in cyber-physical systems (CPS), while highly accurate, are simply pattern-matchers.~With limited security guarantees, there are concerns for their reliability in safety-critical and contested domains. To advance assured AI, we advocate for a paradigm shift that imbues data-driven perception models with symbolic structure, inspired by a human's ability to reason over low-level features and high-level context. We propose a neuro-symbolic paradigm for perception (NeuSPaPer) and illustrate how joint object detection and scene graph generation (SGG) yields deep scene understanding.~Powered by foundation models for offline knowledge extraction and specialized SGG algorithms for real-time deployment, we design a framework leveraging structured relational graphs that ensures the integrity of situational awareness in autonomy. Using physics-based simulators and real-world datasets, we demonstrate how SGG bridges the gap between low-level sensor perception and high-level reasoning, establishing a foundation for resilient, context-aware AI and advancing trusted autonomy in CPS.

Figures

Figures reproduced from arXiv: 2505.21322 by the authors.

Figure 1
Figure 1. Attacker can alter the semantic understanding of the scene while being stealthy to multi [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Neuro-symbolic paradigm for perception performs object detection, classification, and [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Scene pairs with below images. (a) BEV projection of LiDAR point cloud from nuScenes [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: (a) Raw camera image feeds foundation model that (b) directly builds scene graph. Node [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Joint perception and scene graph generation performed using EGTR model from [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: (a) Foundation model jointly detects objects and builds scene graph from image. (b,c) [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Case study of using scene graph generation to secure multi-sensor fusion from attacks on [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Case study of using scene graph generation to secure multi-sensor fusion from attacks on [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 29 canonical work pages

  1. [1]

    Architecting artificial intelligence for autonomous cars: The openpilot framework

    Luciano Baresi and Damian A Tamburri. Architecting artificial intelligence for autonomous cars: The openpilot framework. In European Conference on Software Architecture, pages 189--204. Springer, 2023

  2. [2]

    Geometric deep learning: Grids, groups, graphs, geodesics, and gauges

    Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veli c kovi \'c . Geometric deep learning: Grids, groups, graphs, geodesics, and gauges. arXiv preprint arXiv:2104.13478, 2021

  3. [3]

    nuscenes: A multimodal dataset for autonomous driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11621--11631, 2020

  4. [4]

    Adversarial sensor attack on lidar-based perception in autonomous driving

    Yulong Cao, Chaowei Xiao, Benjamin Cyr, Yimeng Zhou, Won Park, Sara Rampazzi, Qi Alfred Chen, Kevin Fu, and Z Morley Mao. Adversarial sensor attack on lidar-based perception in autonomous driving. In Proceedings of the 2019 ACM SIGSAC conference on computer and communications security, pages 2267--2281, 2019

  5. [5]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In European conference on computer vision, pages 213--229. Springer, 2020

  6. [6]

    Adversarial examples are not easily detected: Bypassing ten detection methods

    Nicholas Carlini and David Wagner. Adversarial examples are not easily detected: Bypassing ten detection methods. In Proceedings of the 10th ACM workshop on artificial intelligence and security, pages 3--14, 2017

  7. [7]

    A review: Knowledge reasoning over knowledge graph

    Xiaojun Chen, Shengbin Jia, and Yang Xiang. A review: Knowledge reasoning over knowledge graph. Expert systems with applications, 141: 0 112948, 2020

  8. [8]

    Scene Graph Generation in Autonomous Driving: a Neuro-symbolic approach

    Paolo Emmanuel Ilario Dimasi. Scene Graph Generation in Autonomous Driving: a Neuro-symbolic approach. PhD thesis, Politecnico di Torino, 2023

Show all 36 references
  1. [9]

    Carla: An open urban driving simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Antonio Lopez, and Vladlen Koltun. Carla: An open urban driving simulator. In Conference on robot learning, pages 1--16. PMLR, 2017

  2. [10]

    Multisensor data fusion

    Hugh Durrant-Whyte and Thomas C Henderson. Multisensor data fusion. Springer handbook of robotics, pages 867--896, 2016

  3. [11]

    Robust physical-world attacks on deep learning visual classification

    Kevin Eykholt, Ivan Evtimov, Earlence Fernandes, Bo Li, Amir Rahmati, Chaowei Xiao, Atul Prakash, Tadayoshi Kohno, and Dawn Song. Robust physical-world attacks on deep learning visual classification. In Proceedings of the IEEE conference on computer vision and pattern recognit...

  4. [12]

    Uav-based multi-sensor data fusion and machine learning algorithm for yield prediction in wheat

    Shuaipeng Fei, Muhammad Adeel Hassan, Yonggui Xiao, Xin Su, Zhen Chen, Qian Cheng, Fuyi Duan, Riqiang Chen, and Yuntao Ma. Uav-based multi-sensor data fusion and machine learning algorithm for yield prediction in wheat. Precision agriculture, 24 0 (1): 0 187--212, 2023

  5. [13]

    Adversarial attacks on medical machine learning

    Samuel G Finlayson, John D Bowers, Joichi Ito, Jonathan L Zittrain, Andrew L Beam, and Isaac S Kohane. Adversarial attacks on medical machine learning. Science, 363 0 (6433): 0 1287--1289, 2019

  6. [14]

    Neuro-symbolic reasoning in the traffic domain

    Leilani H Gilpin and Filip Ilievski. Neuro-symbolic reasoning in the traffic domain. J AI Res, 15 0 (3): 0 123--145, 2021

  7. [15]

    Road-r: the autonomous driving dataset with logical requirements

    Eleonora Giunchiglia, Mihaela C a t a lina Stoian, Salman Khan, Fabio Cuzzolin, and Thomas Lukasiewicz. Road-r: the autonomous driving dataset with logical requirements. Machine Learning, 112 0 (9): 0 3261--3291, 2023

  8. [16]

    Datasets, models, and algorithms for multi-sensor, multi-agent autonomy using avstack

    R Spencer Hallyburton and Miroslav Pajic. Datasets, models, and algorithms for multi-sensor, multi-agent autonomy using avstack. arXiv preprint arXiv:2312.04970, 2023

  9. [17]

    Security analysis of camera-lidar fusion against black-box attacks on autonomous vehicles

    R Spencer Hallyburton, Yupei Liu, Yulong Cao, Z Morley Mao, and Miroslav Pajic. Security analysis of camera-lidar fusion against black-box attacks on autonomous vehicles. In 31st USENIX Security Symposium (USENIX Security 22), pages 1903--1920, 2022

  10. [18]

    Partial-information, longitudinal cyber attacks on lidar in autonomous vehicles

    R Spencer Hallyburton, Qingzhao Zhang, Z Morley Mao, and Miroslav Pajic. Partial-information, longitudinal cyber attacks on lidar in autonomous vehicles. arXiv preprint arXiv:2303.03470, 2023 a

  11. [19]

    Avstack: An open-source, reconfigurable platform for autonomous vehicle development

    Robert Spencer Hallyburton, Shucheng Zhang, and Miroslav Pajic. Avstack: An open-source, reconfigurable platform for autonomous vehicle development. In Proceedings of the ACM/IEEE 14th International Conference on Cyber-Physical Systems (with CPS-IoT Week 2023), pages 209--220, 2023 b

  12. [20]

    Egtr: Extracting graph from transformer for scene graph generation

    Jinbae Im, JeongYeon Nam, Nokyung Park, Hyungmin Lee, and Seunghyun Park. Egtr: Extracting graph from transformer for scene graph generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24229--24238, 2024

  13. [21]

    Certified robustness for top-k predictions against adversarial perturbations via randomized smoothing

    Jinyuan Jia, Xiaoyu Cao, Binghui Wang, and Neil Zhenqiang Gong. Certified robustness for top-k predictions against adversarial perturbations via randomized smoothing. arXiv preprint arXiv:1912.09899, 2019

  14. [22]

    Image retrieval using scene graphs

    Justin Johnson, Ranjay Krishna, Michael Stark, Li-Jia Li, David Shamma, Michael Bernstein, and Li Fei-Fei. Image retrieval using scene graphs. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3668--3678, 2015

  15. [23]

    Vilt: Vision-and-language transformer without convolution or region supervision

    Wonjae Kim, Bokyung Son, and Ildoo Kim. Vilt: Vision-and-language transformer without convolution or region supervision. In International conference on machine learning, pages 5583--5594. PMLR, 2021

  16. [24]

    Pointpillars: Fast encoders for object detection from point clouds

    Alex H Lang, Sourabh Vora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12697--12705, 2019

  17. [25]

    Sgtr: End-to-end scene graph generation with transformer

    Rongjie Li, Songyang Zhang, and Xuming He. Sgtr: End-to-end scene graph generation with transformer. In proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 19486--19496, 2022

  18. [26]

    Visual relationship detection with language priors

    Cewu Lu, Ranjay Krishna, Michael Bernstein, and Li Fei-Fei. Visual relationship detection with language priors. In Computer Vision--ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11--14, 2016, Proceedings, Part I 14, pages 852--869. Springer, 2016

  19. [27]

    Distillation as a defense to adversarial perturbations against deep neural networks

    Nicolas Papernot, Patrick McDaniel, Xi Wu, Somesh Jha, and Ananthram Swami. Distillation as a defense to adversarial perturbations against deep neural networks. In 2016 IEEE symposium on security and privacy (SP), pages 582--597. IEEE, 2016

  20. [28]

    Potential cyberattacks on automated vehicles

    Jonathan Petit and Steven E Shladover. Potential cyberattacks on automated vehicles. IEEE Transactions on Intelligent transportation systems, 16 0 (2): 0 546--556, 2014

  21. [29]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

  22. [30]

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

    J Redmon. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, 2016

  23. [31]

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

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. IEEE transactions on pattern analysis and machine intelligence, 39 0 (6): 0 1137--1149, 2016

  24. [32]

    Universal adversarial training

    Ali Shafahi, Mahyar Najibi, Zheng Xu, John Dickerson, Larry S Davis, and Tom Goldstein. Universal adversarial training. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 5636--5643, 2020

  25. [33]

    Pv-rcnn: Point-voxel feature set abstraction for 3d object detection

    Shaoshuai Shi, Chaoxu Guo, Li Jiang, Zhe Wang, Jianping Shi, Xiaogang Wang, and Hongsheng Li. Pv-rcnn: Point-voxel feature set abstraction for 3d object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10529--10538, 2020

  26. [34]

    Unbiased scene graph generation from biased training

    Kaihua Tang, Yulei Niu, Jianqiang Huang, Jiaxin Shi, and Hanwang Zhang. Unbiased scene graph generation from biased training. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3716--3725, 2020

  27. [35]

    Physically realizable adversarial examples for lidar object detection

    James Tu, Mengye Ren, Sivabalan Manivasagam, Ming Liang, Bin Yang, Richard Du, Frank Cheng, and Raquel Urtasun. Physically realizable adversarial examples for lidar object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13...

  28. [36]

    Neural motifs: Scene graph parsing with global context

    Rowan Zellers, Mark Yatskar, Sam Thomson, and Yejin Choi. Neural motifs: Scene graph parsing with global context. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5831--5840, 2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.