Pith. sign in

REVIEW 4 major objections 4 minor 68 references

Visual Grounding from Event Cameras

T0 review · 4 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper introduces Talk2Event, the first large-scale benchmark for language-driven object grounding built on event-camera data, with 5,567 real driving scenes, 13,458 objects, and 30,690 validated referring expressions decomposed into ap

desk verdict Talk2Event is a plausible first event-based grounding benchmark, but its event-only validity is unproven. read the letter →

arxiv 2509.09584 v1 pith:W5LRVVC6 submitted 2025-09-11 cs.CV cs.RO

Lingdong Kong , Dongyue Lu , Ao Liang , Rong Li , Yuhao Dong , Tianshuai Hu , Lai Xing Ng , Wei Tsang Ooi
show 1 more author
Benoit R. Cottereau
This is my paper
classification cs.CVcs.RO
keywords eventcamerasvisualgroundingreferringexpressionsbenchmarkdrivingscenesattributedecompositionmultimodalperceptionDSEC
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

Talk2Event is the first benchmark to connect event cameras with natural-language grounding. It gives the community 5,567 real driving scenes where each of 13,458 objects is described by multiple referring expressions, and every expression is split into four interpretable attributes: what the object looks like, whether it is moving or still, where it is relative to the viewer, and how it relates to nearby objects. This attribute design lets researchers study not just whether a model can find an object, but which visual or temporal cues it actually used. If the benchmark works as claimed, it opens a new path for language-guided perception in conditions where ordinary cameras fail, such as fast motion, low light, and high dynamic range.

What carries the argument

The central mechanism is the voxelized event representation, a 4D tensor that discretizes the asynchronous event stream into T temporal bins with polarity channels, preserving spatiotemporal dynamics. Around this, the curation pipeline uses two neighboring frames at t0 ± Δt (Δt = 200 ms) as visual context for a vision-language model (Qwen2-VL) to generate three diverse captions per object, then applies a semi-automated fuzzy-matching and LLM-assisted parsing step to decompose each caption into the four attribute labels, followed by human verification and multiple filtering stages (visibility, redundancy, attribute validity). The resulting benchmark supports three complementary evaluation set

What would settle it

Take a random sample of Talk2Event expressions, show human annotators only the voxelized event stream at the reference time (with no RGB frames), and ask whether each attribute label and the bounding box match what an event-only viewer can see; if a substantial fraction of expressions are unverifiable or contradict the event data, the frame-to-event alignment assumption fails and the benchmark's labels are unreliable even though they passed human filtering.

Watch

Extended reading notes

Core claim

The paper's central claim is that event-based visual grounding is a feasible and well-defined task, and that a large-scale, human-validated benchmark can be built from real driving data. Talk2Event compiles 5,567 curated scenes from the DSEC dataset, with 13,458 annotated objects and 30,690 referring expressions generated by prompting a vision-language model with two neighboring RGB frames around a reference time, then refining via attribute decomposition and human verification. Each expression is labeled with four attribute categories — appearance (static properties), status (motion or state), relation-to-viewer (egocentric position), and relation-to-others (spatial and group context) — mak

Load-bearing premise

The referring expressions are generated from two ordinary camera frames just before and after the event snippet, so if the frame-based description describes appearance or motion that the event stream does not actually show at the reference time, every grounding label is off regardless of later human checking.

Editorial extensions

If this is right

  • Models can learn to ground free-form language directly in event voxels, offering a perception channel that stays reliable under motion blur, low light, and high dynamic range where RGB frames degrade.
  • The three evaluation settings (event-only, frame-only, event+frame) provide a controlled protocol to measure how much temporal event information contributes beyond appearance cues.
  • The four attribute labels transform grounding into a compositional task, so an error can be traced to a specific cue such as motion status or spatial relation rather than treated as a single failure.
  • Talk2Event offers a real-data testbed for multimodal, temporally-aware perception, potentially benefiting robotics, autonomous driving, and human-AI interaction in dynamic scenes.

Reading between the lines

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

  • The attribute-decomposition scheme could transfer to other asynchronous or sparse sensing modalities, such as LiDAR or radar, to build language-grounding benchmarks where temporal motion and ego-relations matter as much as appearance.
  • The 'status' attribute implicitly requires the model to perform event-based motion understanding; a model that succeeds on status is effectively doing motion-predicate grounding, a capability the benchmark exposes and measures.
  • Because all data comes from driving scenes, the relational and appearance vocabulary is likely biased toward road users; extending the same protocol to indoor, aerial, or legged-robot event data would test whether the attribute scheme generalizes across domains.
  • A testable extension is to regenerate referring expressions from event-derived optical flow instead of neighboring frames and compare captions; large divergences would reveal which parts of the current descriptions are not actually encoded in the event stream.
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

4 major / 4 minor

Summary. The paper introduces Talk2Event, claimed to be the first large-scale benchmark for language-driven object grounding using event cameras. The dataset is built from the DSEC driving dataset and consists of 5,567 scenes, 13,458 annotated objects, and 30,690 validated referring expressions. Each expression is decomposed into four attribute categories—appearance, status, relation-to-viewer, and relation-to-others—and is paired with voxelized event streams and synchronized frames. The curation pipeline uses Qwen2-VL prompted with two neighboring RGB frames at t0±Δt, followed by LLM-assisted attribute parsing and human verification. The paper defines the grounding task in Section 2.1 and presents the dataset statistics and qualitative examples in Table 1 and Figures 1–3. No baseline experiments or quantitative evaluation are reported.

Significance. If the annotations are reliable and the task is well-posed, Talk2Event would fill a genuine gap: no existing grounding benchmark uses event data, and the four-attribute schema could enable interpretable, compositional evaluation in dynamic scenes. The use of real-world driving data and the explicit multimodal (event, frame, and both) settings are useful features. However, the manuscript currently lacks external verification of the annotation pipeline, does not demonstrate that the event modality actually supports the linguistic attributes, and provides no baseline results; the benchmark's value therefore remains conditional rather than demonstrated.

major comments (4)
  1. [Section 2.1, Eq. (1)] The event-only input E∈R^{2×T×H×W} has channels for polarity only. Yet the attribute schema includes color under Appearance (e.g., 'white color') and static states under Status (e.g., 'stopped'). Section 2.2 generates descriptions from RGB frames, and the human verification step checks captions against frames, not events. Thus a large subset of expressions may be underdetermined in the event-only setting. The paper reports no statistics on how many of the 30,690 expressions depend on event-inobservable attributes. I request a quantitative analysis of event-inferability, or a filtering criterion that removes samples whose distinguishing attributes cannot be observed from events.
  2. [Section 2.2] The curation pipeline is described only at a high level. No prompts, filtering thresholds, annotation-interface details, annotator counts, or inter-annotator agreement figures are given. The 'fuzzy matching and LLM-assisted parsing' stage is not reproducible from the text. For a dataset paper, this is load-bearing: the central claim is that the 30,690 expressions are 'validated', but the reader cannot assess the quality or consistency of the filtering. Please provide complete pipeline details and agreement statistics, and report how many generated captions were discarded at each stage.
  3. [Section 2.2 and Eq. (1)] The relationship between the event voxelization window [ta,tb] and the reference time t0 is never specified. Captions are generated from frames at t0±Δt, but the corresponding event tensor may not cover the same temporal interval or may be misaligned. If the event window is not carefully synchronized with the frames, then the captions can describe content that is not present in the event representation. Please specify how E is constructed relative to t0 and Δt, and provide evidence of correct alignment (e.g., event-rate plots or a manual verification sample).
  4. [Sections 1–3] The paper defines a new task and a benchmark but includes no baseline experiments or even a simple oracle evaluation. There is no train/validation/test split, no evaluation metric (e.g., IoU threshold), and no feasibility study. Without any numerical result, the claim that Talk2Event is a usable benchmark for 'visual grounding from event cameras' is not supported. I recommend adding at least one simple baseline (e.g., event-only, frame-only, and multimodal) and reporting standard grounding metrics, or explicitly deferring models to future work while providing the evaluation protocol.
minor comments (4)
  1. [Table 1] The column labeled 'Sensor' uses 'Static' for frame/RGB-D/LiDAR and 'Dynamic' for event data; this conflates sensor modality with scene dynamics. Consider renaming to 'Modality' or clarifying that 'Dynamic' refers to the sensor output rather than the scene.
  2. [Figure 2] The word clouds are difficult to read in the printed version. If they are intended to support the claim of distinct linguistic patterns, consider providing a table of top-frequency terms per attribute.
  3. [Section 2.2] The sentence 'Each object is described by three distinct captions' should be reconciled with Table 1: 13,458 objects × 3 captions gives 40,374, but the dataset reports 30,690 expressions. Clarify whether some objects have fewer than three captions after filtering.
  4. [References] References [31] and [32] (WildRefer and another) appear to list the same venue and authors; please verify the citation details.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Talk2Event is a constructed dataset resource; its claims do not reduce to their own inputs.

full rationale

Talk2Event is a benchmark construction paper, not a derivation. The central claims—that the dataset contains 5,567 scenes, 13,458 objects, and 30,690 referring expressions with four attribute labels—are descriptive summaries of the curation pipeline (Section 2.2), not predictions derived from fitted parameters or from a self-referential theorem. The captions are generated by Qwen2-VL from neighboring RGB frames and then human-verified; the attributes are assigned by a semi-automated parsing pipeline with human verification. None of these steps define the target result in terms of itself: the benchmark does not 'predict' its own labels, and no fitted parameter is subsequently renamed as a prediction. The paper cites several prior works involving overlapping authors (e.g., Openess [27], EventFly [28], SeeGround [30], FlexEvent [35]), but these citations are contextual support for prior progress in event perception and open-vocabulary recognition, not load-bearing justifications for the benchmark's existence, novelty, or validity. There is no invoked 'uniqueness theorem' from the authors' prior work, no ansatz smuggled in via citation, and no known empirical pattern merely renamed as unification. The concern raised by a skeptical reader—that RGB-generated captions may include attributes such as color or 'stopped' that the event-only voxel representation cannot encode—is a benchmark validity and task-appropriateness issue, not a circularity issue. It does not amount to the paper deriving a conclusion from premises that already contain that conclusion. The paper is self-contained as a dataset contribution, and no circular step can be exhibited with a specific reduction. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the fidelity of inherited DSEC data and on a partially described LLM-plus-human annotation process. No new physical or mathematical entities are introduced; the only design choices are annotation-schema parameters.

free parameters (2)
  • delta t (temporal context) = 200 ms
    Chosen by hand as the offset to the two surrounding frames used by Qwen2-VL; determines whether motion/status captions match the event interval (Sec. 2.2).
  • Captions per object = 3
    A design choice, not fitted; each object receives three distinct captions, affecting redundancy filtering and diversity (Sec. 2.2).
assumptions (4)
  • domain assumption DSEC event streams and RGB frames are temporally synchronized and correctly aligned
    Captions are generated from frames at t0 +/- delta t and used to label events at t0; assumes alignment (Sec. 2.2).
  • domain assumption Qwen2-VL descriptions are semantically grounded in the depicted object and scene
    The entire annotation set is seeded by LLM outputs before human filtering (Sec. 2.2).
  • domain assumption Human verification catches incorrect, ambiguous, or duplicated captions
    Quality is asserted through visibility, redundancy, and attribute filters, but no inter-annotator statistics are reported (Sec. 2.2).
  • ad hoc to paper The four attribute categories are sufficient to represent grounding cues
    Appearance, status, relation-to-viewer, and relation-to-others are introduced as the decomposition; no experiment shows these suffice for compositional grounding (Sec. 2.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Visual Grounding from Event Cameras." pith.science (2026). https://pith.science/paper/W5LRVVC6

@misc{pith2026250909584,
  author       = {Pith},
  title        = {Pith review of: Visual Grounding from Event Cameras},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W5LRVVC6}},
  note         = {Machine review of arXiv:2509.09584}
}
read the original abstract

Event cameras capture changes in brightness with microsecond precision and remain reliable under motion blur and challenging illumination, offering clear advantages for modeling highly dynamic scenes. Yet, their integration with natural language understanding has received little attention, leaving a gap in multimodal perception. To address this, we introduce Talk2Event, the first large-scale benchmark for language-driven object grounding using event data. Built on real-world driving scenarios, Talk2Event comprises 5,567 scenes, 13,458 annotated objects, and more than 30,000 carefully validated referring expressions. Each expression is enriched with four structured attributes -- appearance, status, relation to the viewer, and relation to surrounding objects -- that explicitly capture spatial, temporal, and relational cues. This attribute-centric design supports interpretable and compositional grounding, enabling analysis that moves beyond simple object recognition to contextual reasoning in dynamic environments. We envision Talk2Event as a foundation for advancing multimodal and temporally-aware perception, with applications spanning robotics, human-AI interaction, and so on.

Figures

Figures reproduced from arXiv: 2509.09584 by the authors.

Figure 1
Figure 1. Grounded scene understanding from event cameras. This work introduces Talk2Event, a novel task and dataset for localizing dynamic objects from event streams using natural language descriptions, where each unique object in the scene is defined by four key attributes: ①Appearance, ②Status, ③Relation-to-Viewer, and ④Relation-to-Others. We find that modeling these attributes enables precise, interpretable, and temporall… view at source ↗
Figure 2
Figure 2. Pipeline of dataset curation. We leverage two surrounding frames at t0 ± ∆t to generate context-aware referring expressions of the event stream at t0. Such a description covers key attributes: appearances, motion changes, spatial relations, and interactions. The word clouds shown on the right side highlight distinct linguistic patterns across the four grounding attributes. to output a bounding box bˆ = (x, y, w, h) … view at source ↗
Figure 3
Figure 3. Dataset examples. We provide several event-based visual grounding examples from the Talk2Event dataset, spanning “car”, “truck”, “bus”, and “pedestrian” classes. For more examples and semantic categories, kindly refer to the dataset page. making Talk2Event significantly more verbose than exist￾ing grounding datasets. Attribute-specific word clouds (see [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

68 extracted references · 7 linked inside Pith

  1. [1]

    Referit3d: Neural listeners for fine-grained 3d object identification in real-world scenes

    Panos Achlioptas, Ahmed Abdelreheem, Fei Xia, Mohamed Elhoseiny, and Leonidas Guibas. Referit3d: Neural listeners for fine-grained 3d object identification in real-world scenes. InECCV, pages 422–440. Springer, 2020

    2020

  2. [2]

    Ddd17: End-to-end davis driving dataset

    Jonathan Binas, Daniel Neil, Shih-Chii Liu, and Tobi Del- bruck. Ddd17: End-to-end davis driving dataset. InInterna- tional Conference on Machine Learning Workshops, pages 1–9, 2017

    2017

  3. [3]

    A 240× 180 130 db 3 µs latency global shutter spatiotemporal vision sensor.IEEE Journal of Solid-State Circuits, 49(10):2333–2341, 2014

    Christian Brandli, Raphael Berner, Minhao Yang, Shih-Chii Liu, and Tobi Delbruck. A 240× 180 130 db 3 µs latency global shutter spatiotemporal vision sensor.IEEE Journal of Solid-State Circuits, 49(10):2333–2341, 2014

    2014

  4. [4]

    Low-latency event- based visual odometry

    Andrea Censi and Davide Scaramuzza. Low-latency event- based visual odometry. InICRA, pages 703–710, 2014

    2014

  5. [5]

    Recent event camera innovations: A survey

    Bharatesh Chakravarthi, Aayush Atul Verma, Kostas Dani- ilidis, Cornelia Fermuller, and Yezhou Yang. Recent event camera innovations: A survey. InECCV Workshops. Springer, 2024

    2024

  6. [6]

    Ani Hsieh, Christopher Korpela, Vijay Kumar, Camillo J

    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. InCVPR Workshops, pages 4016–4023, 2023

    2023

  7. [7]

    Chang, and Matthias Nießner

    Dave Zhenyu Chen, Angel X. Chang, and Matthias Nießner. Scanrefer: 3d object localization in rgb-d scans using natural language. InECCV, pages 202–221. Springer, 2020

    2020

  8. [8]

    Ecmd: An event- centric multisensory driving dataset for slam.IEEE Trans- actions on Intelligent Vehicles, 9(1):407–416, 2024

    Peiyu Chen, Weipeng Guan, Feng Huang, Yihan Zhong, Weisong Wen, Li-Ta Hsu, and Peng Lu. Ecmd: An event- centric multisensory driving dataset for slam.IEEE Trans- actions on Intelligent Vehicles, 9(1):407–416, 2024

    2024

Show all 68 references
  1. [9]

    Label-free event-based object recognition via joint learning with image reconstruction from events

    Hoonhee Cho, Hyeonseong Kim, Yujeong Chae, and Kuk- Jin Yoon. Label-free event-based object recognition via joint learning with image reconstruction from events. InICCV, pages 19866–19877, 2023

    2023

  2. [10]

    Hue dataset: High-resolution event and frame sequences for low-light vision

    Burak Ercan, Onur Eker, Aykut Erdem, and Erkut Erdem. Hue dataset: High-resolution event and frame sequences for low-light vision. InECCV Workshops, pages 1–18. Springer, 2024

    2024

  3. [11]

    Thomas Finateu, Atsumi Niwa, Daniel Matolin, Koya Tsuchimoto, Andrea Mascheroni, Etienne Reynaud, Poo- ria Mostafalu, Frederick Brady, Ludovic Chotard, Florian LeGoff, et al. 5.10 a 1280×720 back-illuminated stacked temporal contrast event-based vision sensor with 4.86µm pixel...

    2020

  4. [12]

    Event-based vision: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(1):154–180, 2022

    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):...

    2022

  5. [13]

    Pushing the limits of asynchronous graph-based object detection with event cam- eras.arXiv preprint arXiv:2211.12324, 2022

    Daniel Gehrig and Davide Scaramuzza. Pushing the limits of asynchronous graph-based object detection with event cam- eras.arXiv preprint arXiv:2211.12324, 2022

    2022 arXiv

  6. [14]

    Low-latency auto- motive vision with event cameras.Nature, 629(8014):1034– 1040, 2024

    Daniel Gehrig and Davide Scaramuzza. Low-latency auto- motive vision with event cameras.Nature, 629(8014):1034– 1040, 2024

    2024

  7. [15]

    Derpa- nis, and Davide Scaramuzza

    Daniel Gehrig, Antonio Loquercio, Konstantinos G. Derpa- nis, and Davide Scaramuzza. End-to-end learning of rep- resentations for asynchronous event-based data. InICCV, pages 5633–5643, 2019

    2019

  8. [16]

    Recurrent vision transformers for object detection with event cameras

    Mathias Gehrig and Davide Scaramuzza. Recurrent vision transformers for object detection with event cameras. In CVPR, pages 13884–13893, 2023

    2023

  9. [17]

    Dsec: A stereo event camera dataset for driving scenarios.RA-L, 6(3):4947–4954, 2021

    Mathias Gehrig, Willem Aarents, Daniel Gehrig, and Davide Scaramuzza. Dsec: A stereo event camera dataset for driving scenarios.RA-L, 6(3):4947–4954, 2021

    2021

  10. [18]

    Hierarchical neural memory network for low latency event processing

    Ryuhei Hamaguchi, Yasutaka Furukawa, Masaki Onishi, and Ken Sakurada. Hierarchical neural memory network for low latency event processing. InCVPR, pages 22867–22876, 2023

    2023

  11. [19]

    Evsegsnn: Neuromor- phic semantic segmentation for event data.arXiv preprint arXiv:2406.14178, 2024

    Dalia Hareb and Jean Martinet. Evsegsnn: Neuromor- phic semantic segmentation for event data.arXiv preprint arXiv:2406.14178, 2024

    2024 arXiv

  12. [20]

    Event-aided direct sparse odometry

    Javier Hidalgo-Carri ´o, Guillermo Gallego, and Davide Scaramuzza. Event-aided direct sparse odometry. InCVPR, pages 5781–5790, 2022

    2022

  13. [21]

    Bottom up top down detection transform- ers for language grounding in images and point clouds

    Ayush Jain, Nikolaos Gkanatsios, Ishita Mediratta, and Kate- rina Fragkiadaki. Bottom up top down detection transform- ers for language grounding in images and point clouds. In ECCV, pages 417–433. Springer, 2022

    2022

  14. [22]

    To- wards robust event-based networks for nighttime via un- paired day-to-night event translation

    Yuhwan Jeong, Hoonhee Cho, and Kuk-Jin Yoon. To- wards robust event-based networks for nighttime via un- paired day-to-night event translation. InECCV, pages 286–

  15. [23]

    Efficient learning of event-based dense representation using hierarchical mem- ories with adaptive update

    Uday Kamal and Saibal Mukhopadhyay. Efficient learning of event-based dense representation using hierarchical mem- ories with adaptive update. InECCV, pages 74–89. Springer, 2024

    2024

  16. [24]

    Mdetr- modulated detection for end-to-end multi-modal understand- ing

    Aishwarya Kamath, Mannat Singh, Yann LeCun, Gabriel Synnaeve, Ishan Misra, and Nicolas Carion. Mdetr- modulated detection for end-to-end multi-modal understand- ing. InICCV, pages 1780–1790, 2021

    2021

  17. [25]

    N-imagenet: Towards robust, fine-grained object recognition with event cameras

    Junho Kim, Jaehyeok Bae, Gangin Park, Dongsu Zhang, and Young Min Kim. N-imagenet: Towards robust, fine-grained object recognition with event cameras. InICCV, pages 2146– 2156, 2021

    2021

  18. [26]

    Text2pos: Text-to-point-cloud cross-modal localiza- tion

    Manuel Kolmet, Qunjie Zhou, Aljo ˇsa Oˇsep, and Laura Leal- Taix´e. Text2pos: Text-to-point-cloud cross-modal localiza- tion. InCVPR, pages 6687–6696, 2022

    2022

  19. [27]

    Cot- tereau, and Wei Tsang Ooi

    Lingdong Kong, Youquan Liu, Lai Xing Ng, Benoit R. Cot- tereau, and Wei Tsang Ooi. Openess: Event-based semantic scene understanding with open vocabularies. InCVPR, pages 15686–15698, 2024

    2024

  20. [28]

    Cottereau

    Lingdong Kong, Dongyue Lu, Xiang Xu, Lai Xing Ng, Wei Tsang Ooi, and Benoit R. Cottereau. Eventfly: Event camera perception from ground to the sky. InCVPR, pages 1472–1484, 2025

    2025

  21. [29]

    Geochat: Grounded large vision-language model for remote sensing

    Kartik Kuckreja, Muhammad Sohail Danish, Muzammal Naseer, Abhijit Das, Salman Khan, and Fahad Shahbaz Khan. Geochat: Grounded large vision-language model for remote sensing. InCVPR, pages 27831–27840, 2024. 5

    2024

  22. [30]

    Seeground: See and ground for zero-shot open- vocabulary 3d visual grounding

    Rong Li, Shijie Li, Lingdong Kong, Xulei Yang, and Jun- wei Liang. Seeground: See and ground for zero-shot open- vocabulary 3d visual grounding. InCVPR, pages 3707– 3717, 2025

    2025

  23. [31]

    Wildrefer: 3d object localization in large-scale dynamic scenes with multi-modal visual data and natural language

    Zhenxiang Lin, Xidong Peng, Peishan Cong, Ge Zheng, Yu- jin Sun, Yuenan Hou, Xinge Zhu, Sibei Yang, and Yuexin Ma. Wildrefer: 3d object localization in large-scale dynamic scenes with multi-modal visual data and natural language. In ECCV, pages 456–473. Springer, 2024

    2024

  24. [32]

    Context-aware biaffine localizing network for temporal sentence grounding

    Daizong Liu, Xiaoye Qu, Jianfeng Dong, Pan Zhou, Yu Cheng, Wei Wei, Zichuan Xu, and Yulai Xie. Context-aware biaffine localizing network for temporal sentence grounding. InCVPR, pages 11235–11244, 2021

    2021

  25. [33]

    Explor- ing optical-flow-guided motion and detection-based appear- ance for temporal sentence grounding.IEEE Transactions on Multimedia, 25:8539–8553, 2023

    Daizong Liu, Xiang Fang, Wei Hu, and Pan Zhou. Explor- ing optical-flow-guided motion and detection-based appear- ance for temporal sentence grounding.IEEE Transactions on Multimedia, 25:8539–8553, 2023

    2023

  26. [34]

    A survey on text-guided 3d visual grounding: elements, recent advances, and future directions.arXiv preprint arXiv:2406.05785, 2024

    Daizong Liu, Yang Liu, Wencan Huang, and Wei Hu. A survey on text-guided 3d visual grounding: elements, recent advances, and future directions.arXiv preprint arXiv:2406.05785, 2024

    2024 arXiv

  27. [35]

    Flexevent: Towards flexible event-frame object detection at varying operational frequen- cies.arXiv preprint arXiv:2412.06708, 2024

    Dongyue Lu, Lingdong Kong, Gim Hee Lee, Camille Si- mon Chane, and Wei Tsang Ooi. Flexevent: Towards flexible event-frame object detection at varying operational frequen- cies.arXiv preprint arXiv:2412.06708, 2024

    2024

  28. [36]

    Maqueda, Antonio Loquercio, Guillermo Gallego, Narciso Garc´ıa, and Davide Scaramuzza

    Ana I. Maqueda, Antonio Loquercio, Guillermo Gallego, Narciso Garc´ıa, and Davide Scaramuzza. Event-based vision meets deep learning on steering prediction for self-driving cars. InCVPR, pages 5419–5427, 2018

    2018

  29. [37]

    Cityrefer: geography-aware 3d visual grounding dataset on city-scale point cloud data

    Taiki Miyanishi, Fumiya Kitamori, Shuhei Kurita, Jungdae Lee, Motoaki Kawanabe, and Nakamasa Inoue. Cityrefer: geography-aware 3d visual grounding dataset on city-scale point cloud data. InNeurIPS, 2023

    2023

  30. [38]

    Continuous-time visual-inertial odometry for event cameras.IEEE Transactions on Robotics, 34(6): 1425–1440, 2018

    Elias Mueggler, Guillermo Gallego, Henri Rebecq, and Da- vide Scaramuzza. Continuous-time visual-inertial odometry for event cameras.IEEE Transactions on Robotics, 34(6): 1425–1440, 2018

    2018

  31. [39]

    Learning to detect objects with a 1 megapixel event camera

    Etienne Perot, Pierre De Tournemire, Davide Nitti, Jonathan Masci, and Amos Sironi. Learning to detect objects with a 1 megapixel event camera. InNeurIPS, pages 16639–16652, 2020

    2020

  32. [40]

    Carlos Plou, Nerea Gallego, Alberto Sabater, Eduardo Mon- tijano, Pablo Urcola, Luis Montesano, Ruben Martinez- Cantin, and Ana C. Murillo. Eventsleep: Sleep ac- tivity recognition with event cameras.arXiv preprint arXiv:2404.01801, 2024

    2024 arXiv

  33. [41]

    A qvga 143 db dynamic range frame-free pwm image sensor with lossless pixel-level video compression and time-domain cds.IEEE Journal of Solid-State Circuits, 46(1):259–275, 2010

    Christoph Posch, Daniel Matolin, and Rainer Wohlgenannt. A qvga 143 db dynamic range frame-free pwm image sensor with lossless pixel-level video compression and time-domain cds.IEEE Journal of Solid-State Circuits, 46(1):259–275, 2010

    2010

  34. [42]

    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

    1964

  35. [43]

    4.1 a 640×480 dynamic vision sensor with a 9µm pixel and 300meps address-event representation

    Bongki Son, Yunjae Suh, Sungho Kim, Heejae Jung, Jun- Seok Kim, Changwoo Shin, Keunju Park, Kyoobin Lee, Jin- man Park, Jooyeon Woo, et al. 4.1 a 640×480 dynamic vision sensor with a 9µm pixel and 300meps address-event representation. InIEEE International Solid-State Circuits ...

    2017

  36. [44]

    Neuromor- phic stereo vision: A survey of bio-inspired sensors and al- gorithms.Frontiers in Neuroscience, 13:28, 2019

    Lea Steffen, Daniel Reichard, Jakob Weinland, Jacques Kaiser, Arne Roennau, and R ¨udiger Dillmann. Neuromor- phic stereo vision: A survey of bio-inspired sensors and al- gorithms.Frontiers in Neuroscience, 13:28, 2019

    2019

  37. [45]

    Visual grounding in remote sensing images

    Yuxi Sun, Shanshan Feng, Xutao Li, Yunming Ye, Jian Kang, and Xu Huang. Visual grounding in remote sensing images. InACM MM, pages 404–412, 2022

    2022

  38. [46]

    Ess: Learning event-based semantic seg- mentation from still images

    Zhaoning Sun, Nico Messikommer, Daniel Gehrig, and Da- vide Scaramuzza. Ess: Learning event-based semantic seg- mentation from still images. InECCV, pages 341–357. Springer, 2022

    2022

  39. [47]

    Learning two-branch neural networks for image-text match- ing tasks.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 41(2):394–407, 2018

    Liwei Wang, Yin Li, Jing Huang, and Svetlana Lazebnik. Learning two-branch neural networks for image-text match- ing tasks.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 41(2):394–407, 2018

    2018

  40. [48]

    Neighbourhood watch: Refer- ring expression comprehension via language-guided graph attention networks

    Peng Wang, Qi Wu, Jiewei Cao, Chunhua Shen, Lianli Gao, and Anton van den Hengel. Neighbourhood watch: Refer- ring expression comprehension via language-guided graph attention networks. InCVPR, pages 1960–1968, 2019

    1960

  41. [49]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution.arXiv preprint arXiv:2409.12191, 2024

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Jun- yang Lin. Qwen2-vl: Enhancing vision-language model’s ...

    2024 arXiv

  42. [50]

    Event stream-based visual object tracking: A high-resolution benchmark dataset and a novel baseline

    Xiao Wang, Shiao Wang, Chuanming Tang, Lin Zhu, Bo Jiang, Yonghong Tian, and Jin Tang. Event stream-based visual object tracking: A high-resolution benchmark dataset and a novel baseline. InCVPR, pages 19248–19257, 2024

    2024

  43. [51]

    Referring multi- object tracking

    Dongming Wu, Wencheng Han, Tiancai Wang, Xingping Dong, Xiangyu Zhang, and Jianbing Shen. Referring multi- object tracking. InCVPR, pages 14633–14642, 2023

    2023

  44. [52]

    To- wards open vocabulary learning: A survey.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 46(7): 5092–5113, 2024

    Jianzong Wu, Xiangtai Li, Shilin Xu, Haobo Yuan, Henghui Ding, Yibo Yang, Xia Li, Jiangning Zhang, Yunhai Tong, Xudong Jiang, Bernard Ghanem, and Dacheng Tao. To- wards open vocabulary learning: A survey.IEEE Transac- tions on Pattern Analysis and Machine Intelligence, 46(7): ...

    2024

  45. [53]

    Towards visual grounding: A survey

    Linhui Xiao, Xiaoshan Yang, Xiangyuan Lan, Yaowei Wang, and Changsheng Xu. Towards visual grounding: A survey. arXiv preprint arXiv:2412.20206, 2024

    2024

  46. [54]

    Tubedetr: Spatio-temporal video ground- ing with transformers

    Antoine Yang, Antoine Miech, Josef Sivic, Ivan Laptev, and Cordelia Schmid. Tubedetr: Spatio-temporal video ground- ing with transformers. InCVPR, pages 16442–16453, 2022

    2022

  47. [55]

    Cross-modal re- lationship inference for grounding referring expressions

    Sibei Yang, Guanbin Li, and Yizhou Yu. Cross-modal re- lationship inference for grounding referring expressions. In CVPR, pages 4145–4154, 2019

    2019

  48. [56]

    Event camera data pre- training

    Yan Yang, Liyuan Pan, and Liu Liu. Event camera data pre- training. InICCV, pages 10699–10709, 2023

    2023

  49. [57]

    A fast and accurate one- stage approach to visual grounding

    Zhengyuan Yang, Boqing Gong, Liwei Wang, Wenbing Huang, Dong Yu, and Jiebo Luo. A fast and accurate one- stage approach to visual grounding. InICCV, pages 4683– 4693, 2019. 6

    2019

  50. [58]

    Sat: 2d semantics assisted training for 3d visual grounding

    Zhengyuan Yang, Songyang Zhang, Liwei Wang, and Jiebo Luo. Sat: 2d semantics assisted training for 3d visual grounding. InICCV, pages 1856–1866, 2021

    2021

  51. [59]

    Berg, and Tamara L

    Licheng Yu, Patrick Poirson, Shan Yang, Alexander C. Berg, and Tamara L. Berg. Modeling context in referring expres- sions. InECCV, pages 69–85. Springer, 2016

    2016

  52. [60]

    Visual programming for zero-shot open-vocabulary 3d visual grounding

    Zhihao Yuan, Jinke Ren, Chun-Mei Feng, Hengshuang Zhao, Shuguang Cui, and Zhen Li. Visual programming for zero-shot open-vocabulary 3d visual grounding. InCVPR, pages 20623–20633, 2024

    2024

  53. [61]

    Rsvg: Exploring data and models for visual grounding on remote sensing data

    Yang Zhan, Zhitong Xiong, and Yuan Yuan. Rsvg: Exploring data and models for visual grounding on remote sensing data. IEEE Transactions on Geoscience and Remote Sensing, 61: 1–13, 2023

    2023

  54. [62]

    Mono3dvg: 3d visual grounding in monocular images

    Yang Zhan, Yuan Yuan, and Zhitong Xiong. Mono3dvg: 3d visual grounding in monocular images. InAAAI, pages 6988–6996, 2024

    2024

  55. [63]

    3dvg- transformer: Relation modeling for visual grounding on point clouds

    Lichen Zhao, Daigang Cai, Lu Sheng, and Dong Xu. 3dvg- transformer: Relation modeling for visual grounding on point clouds. InICCV, pages 2928–2937, 2021

    2021

  56. [64]

    Deep learning for event-based vision: A comprehensive survey and bench- marks.arXiv preprint arXiv:2302.08890, 2023

    Xu Zheng, Yexin Liu, Yunfan Lu, Tongyan Hua, Tianbo Pan, Weiming Zhang, Dacheng Tao, and Lin Wang. Deep learning for event-based vision: A comprehensive survey and bench- marks.arXiv preprint arXiv:2302.08890, 2023

    2023 arXiv

  57. [65]

    Ex- act: Language-guided conceptual reasoning and uncertainty estimation for event-based action recognition and more

    Jiazhou Zhou, Xu Zheng, Yuanhuiyi Lyu, and Lin Wang. Ex- act: Language-guided conceptual reasoning and uncertainty estimation for event-based action recognition and more. In CVPR, pages 18633–18643, 2024

    2024

  58. [66]

    Geoground: A unified large vision-language model

    Yue Zhou, Mengcheng Lan, Xiang Li, Yiping Ke, Xue Jiang, Litong Feng, and Wayne Zhang. Geoground: A unified large vision-language model. for remote sensing visual grounding. arXiv preprint arXiv:2411.11904, 2024

    2024 arXiv

  59. [67]

    The multi- vehicle stereo event camera dataset: An event camera dataset for 3d perception.RA-L, 3(3):2032–2039, 2018

    Alex Zihao Zhu, Dinesh Thakur, Tolga ¨Ozaslan, Bernd Pfrommer, Vijay Kumar, and Kostas Daniilidis. The multi- vehicle stereo event camera dataset: An event camera dataset for 3d perception.RA-L, 3(3):2032–2039, 2018

  60. [68]

    Cear: Com- prehensive event camera dataset for rapid perception of agile quadruped robots.RA-L, 9(10):8999–9006, 2024

    Shifan Zhu, Zixun Xiong, and Donghyun Kim. Cear: Com- prehensive event camera dataset for rapid perception of agile quadruped robots.RA-L, 9(10):8999–9006, 2024. 7

    2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.