pith. sign in

arxiv: 2605.20821 · v1 · pith:7FXAWESWnew · submitted 2026-05-20 · 💻 cs.CV · cs.RO

VSCD: Video-based Scene Change Detection in Unaligned Scenes

Jiae Yoon , Ue-Hwan Kim This is my paper

Pith reviewed 2026-05-21 05:08 UTC · model grok-4.3

classification 💻 cs.CV cs.RO
keywords scene change detectionunaligned videopixel-wise maskmulti-reference modelpatch correspondencelong-term autonomymobile robot
0
0 comments X

The pith

A query-centric model detects pixel-wise changes in unsynchronized videos of indoor scenes by aligning local patches and fusing confidence-weighted features.

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

The paper establishes a new setting for detecting scene changes from reference and query videos recorded at different times with completely free camera motion and no temporal synchronization. It constructs a large benchmark dataset containing over 1.1 million frames with pixel-accurate change annotations to support this task. The proposed query-centric multi-reference model learns to match information across time implicitly from the change mask supervision alone. It aligns reference features to the current query frame through local patch correspondences and combines the resulting change features using both frame-level and patch-level confidence measures before generating the final high-resolution change mask. If this works as claimed, robots could monitor and adapt to environmental changes in long-term operations without requiring fixed camera positions or manual video alignment.

Core claim

By training directly on change-mask labels, the model implicitly learns temporal matching. It takes multiple reference frames, aligns their features to the query via local patch correspondence, fuses per-candidate change features with frame-level and patch-level confidence, and decodes a high-resolution pixel-wise change mask for each query frame.

What carries the argument

The query-centric multi-reference model that performs alignment through local patch correspondence and fuses change features using confidence scores at frame and patch levels.

If this is right

  • Outperforms strong image-based and video-based baselines on the introduced benchmark.
  • Transfers to real-world settings as shown by deployment on a mobile robot.
  • Supports two downstream tasks: visual surveillance and object incremental learning.
  • Handles large numbers of appearing and disappearing objects between the reference and query videos.

Where Pith is reading between the lines

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

  • Extending the approach to outdoor or longer-duration videos could test the limits of patch-based alignment under greater appearance variation.
  • Integrating this change detection into broader robotic mapping systems might enable more robust lifelong learning without explicit map updates.
  • The implicit learning from supervision suggests that separate optical flow or registration modules could be avoided in similar video understanding tasks.
  • Evaluating on datasets with significant lighting changes would reveal if the current fusion strategy generalizes beyond the simulated and real test sets provided.

Load-bearing premise

Local patch correspondence between query and reference features produces reliable alignment and change decisions under fully unconstrained camera motion and large temporal gaps.

What would settle it

Running the model on a set of real videos featuring very large viewpoint shifts or time gaps not represented in the training data and checking whether the predicted pixel-wise change masks match human annotations.

Figures

Figures reproduced from arXiv: 2605.20821 by Jiae Yoon, Ue-Hwan Kim.

Figure 1
Figure 1. Figure 1: Comparison between existing change detection datasets and our VSCD benchmark. Unlike prior image-based and video-based datasets that assume similar viewpoints or trajec￾tories, VSCD features unconstrained camera motion, which leads to strong misalignment between reference and query videos. 1. Introduction Embodied agents that operate autonomously in real environ￾ments face a crucial challenge: the world is… view at source ↗
Figure 2
Figure 2. Figure 2: The architecture of the proposed VSCDNet. The model consists of three stages: frame-level alignment to select reference candidates, patch-level correspondence for viewpoint compensation, and confidence-aware fusion followed by query-guided decoding to predict change masks. keeping the encoder frozen, we use a learnable alignment token updated by a lightweight attention-based aggregation head attending to p… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison results on VSCD dataset. The top row shows results on the synthetic data, and the bottom row shows results on the real-world data. 2025). For AOD, we include TCF-LMO (Padilla et al., 2023) and PBCD-MC (Tavares et al., 2025). We do not include methods that require camera pose, geometric calibration, or explicit 3D/multi-view reconstruction as direct baselines, because such inputs and … view at source ↗
Figure 4
Figure 4. Figure 4: Real-world robotic application of video scene change detection. A mobile robot captures reference and query videos under unconstrained motion. VSCD supports (left) visual surveillance by localizing abnormal changes and (right) object incremental learning by highlighting newly introduced objects over time. In contrast, our method produces stable and accurate change masks even under severe viewpoint differen… view at source ↗
Figure 5
Figure 5. Figure 5: Example of generating data in the AI2Thor simulator (Kolve et al., 2017a). Reference and query videos are captured along different trajectories in the same environment with varying object configurations, and pixel-wise change masks are generated by rendering changed objects along the query trajectory. A.1. Synthetic Data Generation Details A.1.1. SIMULATORS AND RENDERING FIDELITY We generate synthetic VSCD… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison results on synthetic VSCD dataset. VSCDNet produces more coherent and complete change masks than representative baselines across both AI2-THOR (top; Mid graphics) and Unreal Engine (bottom; High graphics). We first observe that varying K, the number of top reference frames retrieved per query frame from frame-level matching, has little effect on performance. Using K = 2 or K = 6 yiel… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison results on real-world VSCD dataset. Despite real sensing noise, illumination variation, and unconstrained camera motion, VSCDNet maintains coherent object-level change localization and reduces spurious detections compared to baselines. results in only minor performance differences. This suggests that our candidate construction strategy is robust and does not rely on a large number of… view at source ↗
read the original abstract

Detecting what has changed in an environment is essential for long-term autonomy, yet most change detection settings assume fixed viewpoints, mild misalignment, or only a few changed objects. We introduce Video-based Scene Change Detection (VSCD), which predicts a pixel-wise change mask for each query frame, given a reference and a query RGB video of the same indoor space recorded at different times under unconstrained camera motion. The two videos are not temporally synchronized, and many object instances may appear or disappear. To study this setting, we build a large-scale benchmark with over 1.1 million frames annotated with pixel-accurate change masks, together with a real-world test set for evaluating transfer beyond simulation. We propose a query-centric multi-reference model that learns temporal matching implicitly from change-mask supervision, aligns candidate reference features to the query via local patch correspondence, and fuses per-candidate change features using frame-level and patch-level confidence before decoding a high-resolution mask once per frame. Our approach achieves state-of-the-art performance against strong image- and video-based baselines, and we validate its real-world impact by deploying it on a mobile robot for two downstream applications -- visual surveillance and object incremental learning.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript introduces Video-based Scene Change Detection (VSCD), a task that requires predicting a pixel-wise change mask for each query frame given an unsynchronized reference RGB video and query RGB video of the same indoor scene under fully unconstrained camera motion. It contributes a new large-scale benchmark with over 1.1 million frames annotated with pixel-accurate change masks plus a real-world test set, and proposes a query-centric multi-reference architecture that learns temporal matching implicitly from supervision, aligns reference features to the query via local patch correspondence, fuses per-candidate change features using frame-level and patch-level confidence scores, and decodes a high-resolution mask. The work reports state-of-the-art quantitative results against image- and video-based baselines and demonstrates deployment on a mobile robot for visual surveillance and object incremental learning.

Significance. If the reported performance gains prove robust, the work would meaningfully advance change detection for long-term robot autonomy by relaxing the common assumptions of fixed viewpoints, synchronized capture, and limited object changes. The scale of the introduced benchmark constitutes a substantial community resource, and the real-robot deployment provides direct evidence of downstream utility. The empirical nature of the approach, however, means its significance hinges on whether the local-patch alignment and fusion strategy reliably handles the stated challenges of repetitive textures, specular surfaces, and large viewpoint shifts.

major comments (2)
  1. [§3.2] §3.2 (Alignment and fusion): The central claim that local patch correspondence produces reliable alignment and change decisions rests on an assumption that may not hold under large unconstrained viewpoint shifts and temporal gaps; indoor scenes with repetitive textures can yield ambiguous correspondences, and the manuscript provides no explicit geometric constraints, global optimization, or targeted failure-case analysis to mitigate this risk.
  2. [Table 2] Table 2 and §4.2: The SOTA performance numbers are presented without sufficient ablation on the contribution of the multi-reference fusion (frame-level vs. patch-level confidence) or on the effect of reference video length; without these controls it is difficult to confirm that the gains are attributable to the query-centric design rather than dataset-specific tuning.
minor comments (2)
  1. [Abstract] Abstract and §2: The exact train/validation/test splits and annotation protocol for the 1.1 million frames are not stated, which would improve reproducibility and allow readers to assess potential label noise.
  2. [Figure 3] Figure 3: The qualitative examples would benefit from explicit indication of the magnitude of camera motion and temporal gap between reference and query sequences to illustrate the operating regime.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment of our work's significance and for the detailed feedback. We have carefully considered the major comments and provide point-by-point responses below. Revisions have been made to incorporate additional analyses and ablations to address the concerns raised.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Alignment and fusion): The central claim that local patch correspondence produces reliable alignment and change decisions rests on an assumption that may not hold under large unconstrained viewpoint shifts and temporal gaps; indoor scenes with repetitive textures can yield ambiguous correspondences, and the manuscript provides no explicit geometric constraints, global optimization, or targeted failure-case analysis to mitigate this risk.

    Authors: We thank the referee for highlighting this important consideration. Our approach relies on learning temporal matching implicitly from the change mask supervision rather than explicit geometric constraints. The local patch correspondence is combined with confidence-based fusion to handle potential ambiguities arising from repetitive textures or large viewpoint shifts. We recognize that a more detailed failure-case analysis would be beneficial. Accordingly, we have added a new subsection in the revised manuscript discussing challenging cases, including examples with repetitive patterns and large temporal gaps, along with qualitative results showing how the model performs in these scenarios. We have also clarified in §3.2 that while global optimization is not employed, the multi-reference fusion provides robustness through learned confidence scores. revision: partial

  2. Referee: [Table 2] Table 2 and §4.2: The SOTA performance numbers are presented without sufficient ablation on the contribution of the multi-reference fusion (frame-level vs. patch-level confidence) or on the effect of reference video length; without these controls it is difficult to confirm that the gains are attributable to the query-centric design rather than dataset-specific tuning.

    Authors: We agree that ablations are necessary to isolate the contributions of the proposed components. In the revised manuscript, we have expanded §4.2 and Table 2 with additional ablation studies. Specifically, we compare the full model against variants that disable frame-level confidence, patch-level confidence, or both, demonstrating the incremental benefits of each. Furthermore, we evaluate performance as a function of reference video length, showing that longer references improve results up to a point, consistent with the multi-reference design. These controls support that the performance gains stem from the query-centric architecture rather than tuning alone. revision: yes

Circularity Check

0 steps flagged

Empirical supervised model with no derivation reducing to self-defined inputs or self-citations

full rationale

The paper presents an end-to-end trained query-centric model that learns temporal matching implicitly from pixel-wise change-mask supervision on a newly introduced benchmark of 1.1M frames. Alignment via local patch correspondence and confidence-based fusion are architectural choices trained directly on the target task rather than derived from equations or prior self-citations. SOTA claims and robot deployment results rest on empirical evaluation against external baselines and a real-world test set, with no load-bearing step that renames a fit as a prediction or imports uniqueness from the authors' own prior work.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard deep-learning assumptions for feature extraction and matching plus the domain assumption that patch-level correspondence can compensate for unconstrained motion; no new physical entities or ad-hoc constants are introduced in the abstract.

axioms (2)
  • domain assumption Local patch correspondence between query and reference features produces usable alignment under unconstrained indoor camera motion
    Invoked in the alignment step that precedes feature fusion
  • domain assumption Change-mask supervision is sufficient to train implicit temporal matching without explicit synchronization
    Stated as the learning mechanism in the model description

pith-pipeline@v0.9.0 · 5736 in / 1444 out tokens · 35419 ms · 2026-05-21T05:08:47.712859+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages · 3 internal anchors

  1. [1]

    Intelligence & Robotics , volume=

    Digital twins to embodied artificial intelligence: review and perspective , author=. Intelligence & Robotics , volume=. 2025 , publisher=

    work page 2025

  2. [2]

    The plastic human brain cortex , author=. Annu. Rev. Neurosci. , volume=. 2005 , publisher=

    work page 2005

  3. [3]

    Developmental Behavioral Neuroscience , pages=

    Induction of brain structure by experience: Substrates for cognitive development , author=. Developmental Behavioral Neuroscience , pages=. 2013 , publisher=

    work page 2013

  4. [4]

    2006 , publisher=

    How the body shapes the way we think: a new view of intelligence , author=. 2006 , publisher=

    work page 2006

  5. [5]

    Psychology of Learning and Motivation , volume=

    Catastrophic interference in connectionist networks: The sequential learning problem , author=. Psychology of Learning and Motivation , volume=. 1989 , publisher=

    work page 1989

  6. [6]

    IEEE Transactions on Cognitive and Developmental Systems , volume=

    Cbcl-pr: A cognitively inspired model for class-incremental learning in robotics , author=. IEEE Transactions on Cognitive and Developmental Systems , volume=. 2023 , publisher=

    work page 2023

  7. [7]

    Advances in Neural Information Processing Systems , volume=

    Few-shot continual active learning by a robot , author=. Advances in Neural Information Processing Systems , volume=

    work page

  8. [8]

    2024 IEEE International Conference on Robotics and Automation , pages=

    Interactive continual learning architecture for long-term personalization of home service robots , author=. 2024 IEEE International Conference on Robotics and Automation , pages=. 2024 , organization=

    work page 2024

  9. [9]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Ego4d: Around the world in 3,000 hours of egocentric video , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  10. [10]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Habitat: A platform for embodied ai research , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  11. [11]

    AI2-THOR: An Interactive 3D Environment for Visual AI

    Ai2-thor: An interactive 3d environment for visual ai , author=. arXiv preprint arXiv:1712.05474 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Proceedings of the National Academy of Sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    work page 2017

  13. [13]

    Advances in Neural Information Processing Systems , volume=

    Experience replay for continual learning , author=. Advances in Neural Information Processing Systems , volume=

    work page

  14. [14]

    IEEE Transactions on Image Processing , volume=

    Image change detection algorithms: a systematic survey , author=. IEEE Transactions on Image Processing , volume=. 2005 , publisher=

    work page 2005

  15. [15]

    Change detection from a street image pair using cnn features and superpixel segmentation , booktitle=

    Sakurada, Ken and Okatani, Takayuki , year=. Change detection from a street image pair using cnn features and superpixel segmentation , booktitle=

    work page

  16. [16]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    TransVCL: attention-enhanced video copy localization network with flexible supervision , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  17. [17]

    European Conference on Computer Vision , pages=

    Self-supervised Video Copy Localization with Regional Token Representation , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    work page 2024

  18. [18]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    A deeply supervised attention metric-based network and an open aerial image dataset for remote sensing change detection , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2021 , publisher=

    work page 2021

  19. [19]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Asymmetric siamese networks for semantic change detection in aerial images , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2021 , publisher=

    work page 2021

  20. [20]

    Remote Sensing , volume=

    A spatial-temporal attention-based method and a new dataset for remote sensing image change detection , author=. Remote Sensing , volume=. 2020 , publisher=

    work page 2020

  21. [21]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2018 , publisher=

    work page 2018

  22. [22]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , pages=

    Creating xBD: A dataset for assessing building damage from satellite imagery , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , pages=

    work page

  23. [23]

    Autonomous Robots , volume=

    Street-view change detection with deconvolutional networks , author=. Autonomous Robots , volume=. 2018 , publisher=

    work page 2018

  24. [24]

    IEEE International Conference on Robotics and Automation , pages=

    Weakly supervised silhouette-based semantic scene change detection , author=. IEEE International Conference on Robotics and Automation , pages=. 2020 , organization=

    work page 2020

  25. [25]

    IEEE/RSJ International Conference on Intelligent Robots and Systems , pages=

    Changesim: Towards end-to-end online scene change detection in industrial indoor environments , author=. IEEE/RSJ International Conference on Intelligent Robots and Systems , pages=. 2021 , organization=

    work page 2021

  26. [26]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Remote sensing image change detection with transformers , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2021 , publisher=

    work page 2021

  27. [27]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Joint spatio-temporal modeling for semantic change detection in remote sensing images , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2024 , publisher=

    work page 2024

  28. [28]

    25th IEEE International Conference on Image Processing , pages=

    Fully convolutional siamese networks for change detection , author=. 25th IEEE International Conference on Image Processing , pages=. 2018 , organization=

    work page 2018

  29. [29]

    IEEE Geoscience and Remote Sensing Letters , volume=

    SNUNet-CD: A densely connected Siamese network for change detection of VHR images , author=. IEEE Geoscience and Remote Sensing Letters , volume=. 2021 , publisher=

    work page 2021

  30. [30]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  31. [31]

    Intelligent Multimedia Analysis for Security Applications , pages=

    Content-based video copy detection--a survey , author=. Intelligent Multimedia Analysis for Security Applications , pages=. 2010 , publisher=

    work page 2010

  32. [32]

    International Telecommunications Symposium , pages=

    An annotated video database for abandoned-object detection in a cluttered environment , author=. International Telecommunications Symposium , pages=. 2014 , organization=

    work page 2014

  33. [33]

    IEEE Transactions on Circuits and Systems I: Regular Papers , volume=

    Anomaly detection in moving-camera video sequences using principal subspace analysis , author=. IEEE Transactions on Circuits and Systems I: Regular Papers , volume=. 2017 , publisher=

    work page 2017

  34. [34]

    IEEE Transactions on Image Processing , volume=

    Domain-transformable sparse representation for anomaly detection in moving-camera videos , author=. IEEE Transactions on Image Processing , volume=. 2019 , publisher=

    work page 2019

  35. [35]

    Signal Processing: Image Communication , volume=

    Change detection in moving-camera videos with limited samples using twin-CNN features and learnable morphological operations , author=. Signal Processing: Image Communication , volume=. 2023 , publisher=

    work page 2023

  36. [36]

    IEEE Access , year=

    Pixel-Based Change Detection in Moving-Camera Videos Using Twin Convolutional Features on a Data-Constrained Scenario , author=. IEEE Access , year=

    work page

  37. [37]

    Proceedings of the IEEE International Conference on Computer Vision Workshops , pages=

    Near-duplicate video retrieval with deep metric learning , author=. Proceedings of the IEEE International Conference on Computer Vision Workshops , pages=

    work page

  38. [38]

    arXiv preprint arXiv:2105.14566 , year=

    CNN retrieval based unsupervised metric learning for near-duplicated video retrieval , author=. arXiv preprint arXiv:2105.14566 , year=

    work page arXiv

  39. [39]

    Pattern Recognition , volume=

    Extremely compact video representation for efficient near-duplicates detection , author=. Pattern Recognition , volume=. 2025 , publisher=

    work page 2025

  40. [40]

    arXiv preprint arXiv:2503.16832 , year=

    Joint Self-Supervised Video Alignment and Action Segmentation , author=. arXiv preprint arXiv:2503.16832 , year=

    work page arXiv

  41. [41]

    Proceedings of the 29th ACM International Conference on Multimedia , pages=

    Video similarity and alignment learning on partial video copy detection , author=. Proceedings of the 29th ACM International Conference on Multimedia , pages=

    work page

  42. [42]

    arXiv preprint arXiv:2305.15679 , year=

    A similarity alignment model for video copy segment matching , author=. arXiv preprint arXiv:2305.15679 , year=

    work page arXiv

  43. [43]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Deep high-resolution representation learning for visual recognition , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2020 , publisher=

    work page 2020

  44. [44]

    IEEE Intelligent Vehicles Symposium , pages=

    Dr-tanet: Dynamic receptive temporal attention network for street scene change detection , author=. IEEE Intelligent Vehicles Symposium , pages=

    work page

  45. [45]

    Pattern Recognition , volume=

    How to reduce change detection to semantic segmentation , author=. Pattern Recognition , volume=

    work page

  46. [46]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Zero-shot scene change detection , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  47. [47]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Towards generalizable scene change detection , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  48. [48]

    10th International Workshop on the Analysis of Multitemporal Remote Sensing Images , pages=

    Deep Siamese multi-scale convolutional network for change detection in multi-temporal VHR images , author=. 10th International Workshop on the Analysis of Multitemporal Remote Sensing Images , pages=. 2019 , organization=

    work page 2019

  49. [49]

    International Conference on Learning Representations , volume=

    Simclr: A simple framework for contrastive learning of visual representations , author=. International Conference on Learning Representations , volume=. 2020 , organization=

    work page 2020

  50. [50]

    International Conference on Machine Learning , pages=

    Learning transferable visual models from natural language supervision , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    work page 2021

  51. [51]

    International Conference on Machine Learning , pages=

    Decaf: A deep convolutional activation feature for generic visual recognition , author=. International Conference on Machine Learning , pages=. 2014 , organization=

    work page 2014

  52. [52]

    , author=

    Abrupt visual onsets and selective attention: evidence from visual search. , author=. Journal of Experimental Psychology: Human Perception and Performance , volume=. 1984 , publisher=

    work page 1984

  53. [53]

    , author=

    Visual salience in the change detection paradigm: the special role of object onset. , author=. Journal of Experimental Psychology: Human Perception and Performance , volume=. 2004 , publisher=

    work page 2004

  54. [54]

    Annual Review of Psychology , volume=

    Change detection , author=. Annual Review of Psychology , volume=. 2002 , publisher=

    work page 2002

  55. [55]

    Frontiers in Human Neuroscience , volume=

    Visual mismatch negativity: a mini-review of non-pathological studies with special populations and stimuli , author=. Frontiers in Human Neuroscience , volume=. 2022 , publisher=

    work page 2022

  56. [56]

    Robotics: Science and Systems , pages=

    Dynamic maps for long-term operation of mobile service robots , author=. Robotics: Science and Systems , pages=

    work page

  57. [57]

    IEEE International Conference on Robotics and Automation , pages=

    Long-term 3D map maintenance in dynamic environments , author=. IEEE International Conference on Robotics and Automation , pages=. 2014 , organization=

    work page 2014

  58. [58]

    Sensors , volume=

    Safe and robust map updating for long-term operations in dynamic environments , author=. Sensors , volume=. 2023 , publisher=

    work page 2023

  59. [59]

    International Conference on Machine Learning , pages=

    Vectormapnet: End-to-end vectorized hd map learning , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    work page 2023

  60. [60]

    International Conference on Machine Learning , pages=

    Path planning using neural a* search , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    work page 2021

  61. [61]

    International Conference on Machine Learning , year=

    Palm-e: An embodied multimodal language model , author=. International Conference on Machine Learning , year=

    work page

  62. [62]

    International Conference on Machine Learning , pages=

    End-to-end active object tracking via reinforcement learning , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    work page 2018

  63. [63]

    International Journal of Remote Sensing , volume=

    Change detection techniques , author=. International Journal of Remote Sensing , volume=. 2004 , publisher=

    work page 2004

  64. [64]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Segment anything , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  65. [65]

    Proceedings of the IEEE International Conference on Computer Vision , pages=

    Flownet: Learning optical flow with convolutional networks , author=. Proceedings of the IEEE International Conference on Computer Vision , pages=

    work page

  66. [66]

    SAM 2: Segment Anything in Images and Videos

    SAM 2: Segment Anything in Images and Videos , author=. arXiv preprint arXiv:2408.00714 , url=

    work page internal anchor Pith review Pith/arXiv arXiv