EVIS: A Physics-Grounded Event Camera Plugin for NVIDIA Isaac Sim

Linli Shi; Ruijun Zhang; Ziyun Wang

EVIS: A Physics-Grounded Event Camera Plugin for NVIDIA Isaac Sim

T0 review · 0 major / 5 minor · reviewed 2026-07-10 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A physics simulator plugin generates labeled event-camera streams that pretrained networks can use without fine-tuning, at real-time rates on one GPU.

desk verdict Solid systems paper that finally puts a usable, real-time event camera inside Isaac Sim; the claim holds and the soft spots are scoped, not fatal. read the letter →

arxiv 2607.08098 v1 pith:SPJ2TFN7 submitted 2026-07-09 cs.CV cs.RO

Linli Shi , Ruijun Zhang , Ziyun Wang This is my paper · ORCID

classification cs.CVcs.RO

keywords eventcamerassimulationphysicssimulatormotion-vectorinterpolationrobotlearningopticalflowevent-basedvision

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

Event cameras report brightness changes asynchronously and are valuable for fast robotics, but labeled recordings for a given robot and scene are scarce and costly. This paper presents EVIS, a plugin that turns any camera in a modern GPU physics simulator into an event camera, so events, rigid-body physics, and perfect ground truth are produced together. The core is a log-intensity contrast model with per-pixel asynchronous reference updates that mirrors real sensors, plus optional noise and motion blur. To keep generation real-time, the system renders only sparse keyframes and fills the gaps by warping with the renderer’s own motion vectors. Frozen networks trained on real event data reconstruct intensity, estimate optical flow, and match features across views on these streams with only gentle degradation as interpolation grows more aggressive. The claim is that this closes the data bottleneck so event-based perception and control can be trained at the same scale as ordinary RGB pipelines.

What carries the argument

The log-intensity contrast event model with per-pixel asynchronous reference latching, accelerated by bidirectional motion-vector warping that synthesizes intermediate frames from sparse keyframes under a constant-velocity assumption.

What would settle it

Run the same pretrained reconstruction, flow, and matching networks on paired real and simulated event streams of identical robot motions and scenes; large, systematic drops in SSIM, endpoint error, or matching precision on the simulated streams would falsify the claim of direct usability.

Watch

Extended reading notes

Core claim

A drop-in event-camera plugin for a GPU physics simulator can generate high-rate, fully labeled event streams that are physics-consistent with the scene, real-time on a single GPU via motion-vector interpolation, and directly usable by frozen pretrained event networks without adaptation.

Load-bearing premise

That warping frames under constant velocity and a simple log-contrast threshold yields events close enough to real cameras for networks trained on real data to work without fine-tuning.

Editorial extensions

If this is right

Any existing physics scene can produce synchronized events, physics, and ground truth by changing only the camera configuration.
Event-based perception and control can be trained at the same batch scale as RGB pipelines inside one simulator.
Real-time generation on a single GPU becomes practical by trading sparse keyframe rendering for motion-vector interpolation.
Optional noise and blur models let users close the remaining gap to hardware without leaving the simulator.

Reading between the lines

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

The same warping pipeline could later support differentiable event generation for end-to-end policy training through the event stream.
Because residual banding appears under occlusion and rotation, the method is most reliable for motions that are mostly translational or moderate in speed relative to the keyframe rate.
Parallel multi-environment generation already amortizes rendering cost, so large-scale multi-robot event datasets become feasible without new hardware.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

This is the tooling paper the event-robotics crowd has needed. EVIS drops a proper asynchronous log-intensity contrast model (per-pixel fire-and-latch, Eqs. 1–3) into Isaac Sim/Lab so any physics scene immediately produces labeled events with frame-perfect ground truth. The practical novelty is the bidirectional motion-vector warp that lets them render sparse keyframes and still hit kHz-rate streams in real time on one GPU (Table 1: 30 imes8 already faster than real time). Code is public.

What they do well is keep the evaluation honest. They freeze pretrained E2VID, E-RAFT and Match-Any-Events and show the streams are usable zero-shot; flow stays sub-pixel EPE even under aggressive interpolation, reconstruction and matching degrade gracefully (Tables 2–4). Residual banding under occlusion/rotation is shown in Fig. 5 and correctly attributed to the constant-velocity assumption and missing source pixels. Noise and motion-blur models are optional and standard. No circular metrics: everything is scored against simulator motion vectors, poses or external frozen nets.

Soft spots are real but proportionate. There is no quantitative real-to-sim sensor match, so the “close enough for transfer” claim rests on the zero-shot network results and visual spacetime clouds rather than event-rate or contrast statistics against a DAVIS or Prophesee. The warp’s linear-motion assumption is acknowledged; they simply recommend raising the base render rate when motion is fast or heavily occluding. Free parameters (C, noise rates, K) are declared as config, not fitted and then sold as predictions. Citation pattern is clean: ESIM, v2e, Gazebo DVS, EvDNeRF are all there; they correctly note none of them sit inside a modern GPU-parallel physics stack.

This is for people who train event perception or control on robots and already live in Isaac Lab. It does not invent a new event model; it makes the existing one usable at scale with physics. I would send it to peer review without hesitation—methods contribution of this scope deserves referee time. Worth reading and worth citing if you need synthetic event data for robot learning.

Referee Report

0 major / 5 minor

Summary. EVIS is a plugin for NVIDIA Isaac Sim / Isaac Lab that turns any RGB camera configuration into a physics-consistent event camera. It implements a standard log-intensity contrast model with per-pixel asynchronous reference latching (Eqs. 1–3), optional sensor non-idealities (threshold mismatch, refractory period, leak/shot/hot-pixel noise, finite bandwidth) and motion blur, and a bidirectional motion-vector warping pipeline (Eqs. 4–6) that synthesizes intermediate frames from sparse RTX keyframes so that high-rate event streams can be generated in real time on a single GPU. Events inherit the simulator’s physics and frame-perfect ground truth and are written as (x,y,t,p) streams. The authors demonstrate real-time throughput (Table 1), zero-shot transfer of frozen pretrained E2VID, E-RAFT and Match-Any-Events models (Tables 2–4, Fig. 3), and qualitative spacetime geometry under several motions and noise settings (Figs. 1, 5).

Significance. Labeled event data for specific robot embodiments and scenes remain scarce; a drop-in, physics-grounded event camera inside a modern GPU-parallel robotics simulator is therefore a practically useful systems contribution. The work is concrete: the event model and warp are fully specified, timing is measured on commodity hardware, and usability is shown by running three independent pretrained networks without fine-tuning. Public code further raises the value for the community. Residual limitations (constant-velocity warp under occlusion/rotation, lack of quantitative real-to-sim sensor matching) are acknowledged and do not negate the tooling claim.

minor comments (5)

The abstract and introduction claim that optional noise and motion blur “further narrow the gap to real cameras,” yet no quantitative real-to-sim comparison (event-rate statistics, contrast-threshold histograms, or network transfer from real recordings) is reported. A short paragraph or appendix acknowledging this as future work would keep the claim proportionate.
Section 4.1: the reuse of the previous-step motion field m_A as A’s forward velocity is an approximation; a one-sentence note on when this breaks (e.g., sudden acceleration) would help readers choose keyframe rates.
Figure 5 caption and surrounding text correctly note residual banding under occlusion and rotation; adding a brief quantitative measure (e.g., event-density variance along the time axis) would make the residual effect easier to compare across configs.
Table 1 reports single-environment timings; a short remark on multi-environment scaling (already mentioned in the text) would strengthen the real-time claim for typical Isaac Lab training loops.
Minor presentation: “V oltmeter” and “DVS-V oltmeter” appear with a space in the related-work section; “spikelab- jhu” in the abstract URL has a space; consistent capitalization of “ON/OFF” would improve polish.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: systems plugin with standard event model, independent simulator GT, and frozen external networks.

full rationale

EVIS is a systems/tooling paper. The event generation chain (log-intensity contrast with per-pixel latch, Eqs. 1–3; optional noise/blur) is the standard hardware model, not derived from the evaluation targets. Motion-vector interpolation (Eqs. 4–6) is an efficiency construction whose constant-velocity and occlusion limits are stated openly (Fig. 5 banding). Downstream metrics compare frozen pretrained networks (E2VID, E-RAFT, Match-Any-Events) against independent simulator ground truth (luminance, renderer motion vectors, relative pose); free parameters (C, noise rates) are declared as configurable sensor settings, not fitted then re-presented as predictions. The paper notes that flow preservation under warp is “expected by construction,” which is an honest design remark rather than a hidden reduction of a claimed prediction. Self-citations (e.g. Match-Any-Events) appear only as one of three evaluation tools, not as load-bearing uniqueness theorems or smuggled ansatze. No step reduces a claimed first-principles result to its own inputs by definition or fit.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

As a systems paper the load-bearing content is engineering choices and empirical validation rather than new physical postulates. Free parameters are the configurable sensor knobs; axioms are the standard event-camera contrast model and the constant-velocity warping assumption. No new physical entities are invented.

free parameters (4)

contrast threshold C = 0.15
Default C=0.15 controls event density; chosen by hand as a typical real-camera value and held fixed in all reported experiments.
noise rates (r_ℓ, r_s, r_h, τ_r, σ_θ, f_c)
Optional sensor non-ideality parameters sampled or set by the user; values affect realism but are not fitted to a target metric in the paper.
interpolation factor K and keyframe rate f_k
User-chosen trade-off between render cost and temporal fidelity; several discrete configs are timed and evaluated.
softmax-splatting β and over-render margin = β=12
β=12 and optional 50-pixel margin are fixed engineering constants that control hole-filling and border quality.

assumptions (3)

domain assumption Events are generated by a per-pixel log-intensity contrast threshold with asynchronous reference latching (standard DVS model).
Invoked in Section 3, Eqs. 1–3; treated as faithful to real hardware without new derivation.
ad hoc to paper Linear motion-vector interpolation between keyframes preserves the true optical-flow field sufficiently for event generation under moderate speeds and limited occlusion.
Core of Section 4; residual banding under rotation/occlusion is acknowledged in Section 6.3 but assumed mild enough for downstream usability.
domain assumption Isaac Sim’s HDR color buffer (linear intensity) and motion vectors are accurate enough to serve as ground-truth inputs for the event model.
Implicit throughout; the plugin inherits the simulator’s rendering and physics without independent validation of those quantities.

how reviews work

0 comments

Cite this review

Pith. "Pith review of EVIS: A Physics-Grounded Event Camera Plugin for NVIDIA Isaac Sim." pith.science (2026). https://pith.science/paper/SPJ2TFN7

@misc{pith2026260708098,
  author       = {Pith},
  title        = {Pith review of: EVIS: A Physics-Grounded Event Camera Plugin for NVIDIA Isaac Sim},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SPJ2TFN7}},
  note         = {Machine review of arXiv:2607.08098}
}

read the original abstract

Event cameras offer microsecond temporal resolution, low latency, and high dynamic range, making them attractive for robotics. However, labeled event-camera data for a specific robot and scene is scarce and expensive to collect, which slows the development of event-based perception and control. We present EVIS: a physics-grounded event camera plugin for NVIDIA Isaac Sim that generates high-rate, fully labeled event streams directly inside a physics simulator. The plugin implements a faithful log-intensity contrast event model with per-pixel asynchronous reference updates; it migrates from a normal RGB camera with few changes and integrates into any Isaac Sim / Isaac Lab scene, inheriting the simulator's physics and frame-perfect ground truth. It is fully configurable, and offers an interpolation option that renders only sparse keyframes and synthesizes the in-between frames through bidirectional motion-vector warping, making real-time generation on a single GPU possible. Optional sensor noise and motion blur further narrow the gap to real cameras. The generated streams are directly usable by pretrained event networks for downstream tasks. Code repository: https://github.com/spikelab-jhu/isaac-sim-event-camera-plugin

Figures

Figures reproduced from arXiv: 2607.08098 by the authors.

**Figure 1.** Top: a Franka wrist-mounted event camera circles a mustard bottle. Middle: a head-mounted event camera on a walking ANYmal [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Motion blur with a 20 ms exposure. Left: sharp 1000 Hz frames and their 1 ms events. Right: the same instant with the exposure enabled; one batch of events covers the whole 20 ms of motion, smearing edges into wide bands. 5.2. Motion blur A real frame camera integrates light over its shutter time, producing motion blur. We model this by averaging the fine config render warp event total vs. real-time 1000 (no warp) 1… view at source ↗

**Figure 3.** Qualitative overview of the downstream-task benchmark at one time instant. Columns: the full-render reference ( [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Event-to-event feature matching. Top: the RGB view from each of the two cameras, which converge on the same objects from a 20◦ separation. Bottom: the event streams of the two views(ON red, OFF blue) with the correspondences found by the pretrained matcher drawn as co…

**Figure 5.** Figure 5: Event point clouds across motions, temporal resolutions, and sensor noise. Each point is a single event in space time (x, y, t), colored by timestamp from blue (start of the clip) to red (end). A power drill undergoes four canonical motions (columns): free fall, out-of…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

EsaacSim: A Multimodal Event Camera Add-on for NVIDIA Isaac Sim
cs.RO 2026-08 conditional novelty 6.0 of 10

EsaacSim is an Isaac Sim add-on that generates online grayscale and Bayer RGGB event streams at up to 960 Hz effective rate, with synchronized RGB, APS, depth, and IMU outputs via ROS 2.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · cited by 1 Pith paper

[1]

Evdnerf: Reconstructing event data with dynamic neural ra- diance fields

Anish Bhattacharya, Ratnesh Madaan, Fernando Cladera, Sai Vemprala, Rogerio Bonatti, Kostas Daniilidis, Ashish Kapoor, Vijay Kumar, Nikolai Matni, and Jayesh K Gupta. Evdnerf: Reconstructing event data with dynamic neural ra- diance fields. InProceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision, pages 5846– 5855, 2024. 2

work page 2024
[2]

Pix2nvs: Parameterized conversion of pixel-domain video frames to neuromorphic vision streams

Yin Bi and Yiannis Andreopoulos. Pix2nvs: Parameterized conversion of pixel-domain video frames to neuromorphic vision streams. In2017 IEEE International Conference on Image Processing (ICIP), pages 1990–1994. IEEE, 2017. 2

work page 1990
[3]

OctoSense: Self-Supervised Learning for Multimodal Robot Perception

Anthony Bisulco, Jeremy Wang, Kostas Daniilidis, Ran- dall Balestriero, and Pratik Chaudhari. Octosense: Self- supervised learning for multimodal robot perception.arXiv preprint arXiv:2606.27317, 2026. 2

work page Pith review arXiv 2026
[4]

M3ed: Multi-robot, multi-sensor, multi-environment event dataset

Kenneth Chaney, Fernando Cladera, Ziyun Wang, Anthony Bisulco, M Ani Hsieh, Christopher Korpela, Vijay Kumar, Camillo J Taylor, and Kostas Daniilidis. M3ed: Multi-robot, multi-sensor, multi-environment event dataset. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4016–4023, 2023. 2

work page 2023
[5]

Event-based vision: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(1):154–180, 2020

Guillermo Gallego, Tobi Delbr ¨uck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J Davison, J ¨org Conradt, Kostas Daniilidis, et al. Event-based vision: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(1):154–180, 2020. 2, 3

work page 2020
[6]

Video to events: Recycling video datasets for event cameras

Daniel Gehrig, Mathias Gehrig, Javier Hidalgo-Carri ´o, and Davide Scaramuzza. Video to events: Recycling video datasets for event cameras. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3586–3595, 2020. 2

work page 2020
[7]

E-raft: Dense optical flow from event cam- eras

Mathias Gehrig, Mario Millh ¨ausler, Daniel Gehrig, and Da- vide Scaramuzza. E-raft: Dense optical flow from event cam- eras. In2021 International Conference on 3D Vision (3DV), pages 197–206. IEEE, 2021. 2, 6

work page 2021
[8]

How to learn a domain-adaptive event simulator? InProceedings of the 29th ACM international conference on multimedia, pages 1275–1283, 2021

Daxin Gu, Jia Li, Yu Zhang, and Yonghong Tian. How to learn a domain-adaptive event simulator? InProceedings of the 29th ACM international conference on multimedia, pages 1275–1283, 2021. 2

work page 2021

Show all 26 references

[9]

Etap: Event- based tracking of any point

Friedhelm Hamann, Daniel Gehrig, Filbert Febryanto, Kostas Daniilidis, and Guillermo Gallego. Etap: Event- based tracking of any point. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 27186– 27196, 2025. 2

2025
[10]

v2e: From video frames to realistic dvs events

Yuhuang Hu, Shih-Chii Liu, and Tobi Delbruck. v2e: From video frames to realistic dvs events. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1312–1321, 2021. 2, 3, 4

2021
[11]

Towards a framework for end-to-end con- trol of a simulated vehicle with spiking neural networks

Jacques Kaiser, J Camilo Vasquez Tieck, Christian Hub- schneider, Peter Wolf, Michael Weber, Michael Hoff, Alexander Friedrich, Konrad Wojtasik, Arne Roennau, Ralf Kohlhaas, et al. Towards a framework for end-to-end con- trol of a simulated vehicle with spiking neural networks...

2016
[12]

Dvs- voltmeter: Stochastic process-based event simulator for dy- namic vision sensors

Songnan Lin, Ye Ma, Zhenhua Guo, and Bihan Wen. Dvs- voltmeter: Stochastic process-based event simulator for dy- namic vision sensors. InEuropean Conference on Computer Vision, pages 578–593. Springer, 2022. 2

2022
[13]

Orbit: A unified simulation framework for interactive robot learning environments.IEEE Robotics and Automation Letters, 8(6):3740–3747, 2023

Mayank Mittal, Calvin Yu, Qinxi Yu, Jingzhou Liu, Nikita Rudin, David Hoeller, Jia Lin Yuan, Ritvik Singh, Yunrong Guo, Hammad Mazhar, et al. Orbit: A unified simulation framework for interactive robot learning environments.IEEE Robotics and Automation Letters, 8(6):3740–3747, 2023. 2

2023
[14]

The event-camera dataset and simulator: Event-based data for pose estimation, visual odometry, and slam.The International journal of robotics research, 36(2):142–149, 2017

Elias Mueggler, Henri Rebecq, Guillermo Gallego, Tobi Del- bruck, and Davide Scaramuzza. The event-camera dataset and simulator: Event-based data for pose estimation, visual odometry, and slam.The International journal of robotics research, 36(2):142–149, 2017. 2 8

2017
[15]

Differentiable event stream simulator for non-rigid 3d track- ing

Jalees Nehvi, Vladislav Golyanik, Franziska Mueller, Hans- Peter Seidel, Mohamed Elgharib, and Christian Theobalt. Differentiable event stream simulator for non-rigid 3d track- ing. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 1302–1311,
[16]

Softmax splatting for video frame interpolation

Simon Niklaus and Feng Liu. Softmax splatting for video frame interpolation. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 5437–5446, 2020. 3

2020
[17]

Text-to- events: Synthetic event camera streams from conditional text input

Joachim Ott, Zuowen Wang, and Shih-Chii Liu. Text-to- events: Synthetic event camera streams from conditional text input. In2024 Neuro Inspired Computational Elements Con- ference (NICE), pages 1–10. IEEE, 2024. 2

2024
[18]

Esim: an open event camera simulator

Henri Rebecq, Daniel Gehrig, and Davide Scaramuzza. Esim: an open event camera simulator. InConference on robot learning, pages 969–982. PMLR, 2018. 2, 3

2018
[19]

High speed and high dynamic range video with an event camera.IEEE transactions on pattern analysis and machine intelligence, 43(6):1964–1980, 2019

Henri Rebecq, Ren ´e Ranftl, Vladlen Koltun, and Davide Scaramuzza. High speed and high dynamic range video with an event camera.IEEE transactions on pattern analysis and machine intelligence, 43(6):1964–1980, 2019. 2, 5

1964
[20]

Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004. 5

2004
[21]

Evac3d: From event-based apparent contours to 3d models via continuous visual hulls

Ziyun Wang, Kenneth Chaney, and Kostas Daniilidis. Evac3d: From event-based apparent contours to 3d models via continuous visual hulls. InEuropean conference on com- puter vision, pages 284–299. Springer, 2022. 2

2022
[22]

Ev-catcher: High-speed object catching using low-latency event-based neural net- works.IEEE Robotics and Automation Letters, 7(4):8737– 8744, 2022

Ziyun Wang, Fernando Cladera, Anthony Bisulco, Daewon Lee, Camillo J Taylor, Kostas Daniilidis, M Ani Hsieh, Daniel D Lee, and V olkan Isler. Ev-catcher: High-speed object catching using low-latency event-based neural net- works.IEEE Robotics and Automation Letters, 7(4):8737–...

2022
[23]

Un- evimo: Unsupervised event-based independent motion seg- mentation

Ziyun Wang, Jinyuan Guo, and Kostas Daniilidis. Un- evimo: Unsupervised event-based independent motion seg- mentation. InEuropean Conference on Computer Vision, pages 228–245. Springer, 2024. 2

2024
[24]

The unreasonable effectiveness of deep features as a perceptual metric

Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 5

2018
[25]

Match-any-events: Zero-shot motion-robust feature match- ing across wide baselines for event cameras.arXiv preprint arXiv:2604.18744, 2026

Ruijun Zhang, Hang Su, Kostas Daniilidis, and Ziyun Wang. Match-any-events: Zero-shot motion-robust feature match- ing across wide baselines for event cameras.arXiv preprint arXiv:2604.18744, 2026. 2, 6

2026 arXiv
[26]

V2ce: Video to continuous events simulator

Zhongyang Zhang, Shuyang Cui, Kaidong Chai, Haowen Yu, Subhasis Dasgupta, Upal Mahbub, and Tauhidur Rah- man. V2ce: Video to continuous events simulator. In2024 IEEE international conference on robotics and automation (ICRA), pages 12455–12461. IEEE, 2024. 2 9

2024

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.

[1] [1]

Evdnerf: Reconstructing event data with dynamic neural ra- diance fields

Anish Bhattacharya, Ratnesh Madaan, Fernando Cladera, Sai Vemprala, Rogerio Bonatti, Kostas Daniilidis, Ashish Kapoor, Vijay Kumar, Nikolai Matni, and Jayesh K Gupta. Evdnerf: Reconstructing event data with dynamic neural ra- diance fields. InProceedings of the IEEE/CVF Winter Con- ference on Applications of Computer Vision, pages 5846– 5855, 2024. 2

work page 2024

[2] [2]

Pix2nvs: Parameterized conversion of pixel-domain video frames to neuromorphic vision streams

Yin Bi and Yiannis Andreopoulos. Pix2nvs: Parameterized conversion of pixel-domain video frames to neuromorphic vision streams. In2017 IEEE International Conference on Image Processing (ICIP), pages 1990–1994. IEEE, 2017. 2

work page 1990

[3] [3]

OctoSense: Self-Supervised Learning for Multimodal Robot Perception

Anthony Bisulco, Jeremy Wang, Kostas Daniilidis, Ran- dall Balestriero, and Pratik Chaudhari. Octosense: Self- supervised learning for multimodal robot perception.arXiv preprint arXiv:2606.27317, 2026. 2

work page Pith review arXiv 2026

[4] [4]

M3ed: Multi-robot, multi-sensor, multi-environment event dataset

Kenneth Chaney, Fernando Cladera, Ziyun Wang, Anthony Bisulco, M Ani Hsieh, Christopher Korpela, Vijay Kumar, Camillo J Taylor, and Kostas Daniilidis. M3ed: Multi-robot, multi-sensor, multi-environment event dataset. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4016–4023, 2023. 2

work page 2023

[5] [5]

Event-based vision: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(1):154–180, 2020

Guillermo Gallego, Tobi Delbr ¨uck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J Davison, J ¨org Conradt, Kostas Daniilidis, et al. Event-based vision: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(1):154–180, 2020. 2, 3

work page 2020

[6] [6]

Video to events: Recycling video datasets for event cameras

Daniel Gehrig, Mathias Gehrig, Javier Hidalgo-Carri ´o, and Davide Scaramuzza. Video to events: Recycling video datasets for event cameras. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3586–3595, 2020. 2

work page 2020

[7] [7]

E-raft: Dense optical flow from event cam- eras

Mathias Gehrig, Mario Millh ¨ausler, Daniel Gehrig, and Da- vide Scaramuzza. E-raft: Dense optical flow from event cam- eras. In2021 International Conference on 3D Vision (3DV), pages 197–206. IEEE, 2021. 2, 6

work page 2021

[8] [8]

How to learn a domain-adaptive event simulator? InProceedings of the 29th ACM international conference on multimedia, pages 1275–1283, 2021

Daxin Gu, Jia Li, Yu Zhang, and Yonghong Tian. How to learn a domain-adaptive event simulator? InProceedings of the 29th ACM international conference on multimedia, pages 1275–1283, 2021. 2

work page 2021