pith. sign in

arxiv: 2605.17584 · v1 · pith:CYLKMREPnew · submitted 2026-05-11 · 💻 cs.CV

VVitCutLER: Towards Unsupervised Object Detection and Segmentation in Videos

Zhijing Lu , Khurram Azeem Hashmi , Didier Stricker , Muhammad Zeshan Afzal This is my paper

Pith reviewed 2026-05-20 22:26 UTC · model grok-4.3

classification 💻 cs.CV
keywords unsupervised video object detectioninstance segmentationpseudo-label generationtemporal consistencycross-frame aggregationVitCutvideo benchmarks
0
0 comments X

The pith

Enforcing cross-frame region consistency in pseudo-labels stabilizes unsupervised video object detection and segmentation.

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

The paper presents VVitCutLER as an unsupervised framework that tackles temporal drift and flickering in video pseudo-labels caused by motion blur, occlusions, and fast dynamics. Its main proposal is VitCut, a pseudo-label generator that maintains stability by enforcing region consistency across frames and includes a distillation decoder for producing instance masks. VVitCutLER builds on this by adding cross-frame feature aggregation to increase overall robustness at the video level. Experiments on standard benchmarks show gains in detection and segmentation accuracy alongside lower temporal instability, underscoring the value of consistent supervision for pixel-level video tasks.

Core claim

VitCut generates temporally stable pseudo-labels by enforcing cross-frame region consistency to limit error accumulation during field degradation, while a distillation decoder handles instance mask prediction; VVitCutLER then layers cross-frame feature aggregation on top to boost video-level robustness, yielding higher detection and segmentation performance with reduced temporal instability on standard benchmarks.

What carries the argument

VitCut, a pseudo-label generator that reduces error accumulation via cross-frame region consistency and uses a distillation decoder for instance mask prediction.

If this is right

  • Detection and segmentation accuracy rises on video benchmarks when temporal consistency is added to pseudo-label generation.
  • Flickering and drift in pseudo-labels decrease because region consistency is maintained across adjacent frames.
  • Video-level robustness improves through the addition of cross-frame feature aggregation after VitCut.
  • Unsupervised pixel-level understanding becomes more practical in real-world settings with motion blur and occlusions.

Where Pith is reading between the lines

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

  • Similar consistency mechanisms could be tested on unsupervised video tracking or action recognition to check whether they reduce drift in those tasks as well.
  • The framework might lower reliance on manual labels for applications such as traffic monitoring or robotic navigation if the stability gains hold across diverse datasets.
  • Extending the cross-frame aggregation to longer sequences or multi-camera setups could reveal whether the same principles scale to more complex video environments.

Load-bearing premise

Enforcing cross-frame region consistency in the pseudo-label generator will reliably cut error accumulation and temporal drift without creating new biases that lower overall performance.

What would settle it

Running the method on a video benchmark where detection and segmentation scores match or fall below non-consistent baselines, or where temporal flicker increases rather than decreases, would show the central claim does not hold.

Figures

Figures reproduced from arXiv: 2605.17584 by Didier Stricker, Khurram Azeem Hashmi, Muhammad Zeshan Afzal, Zhijing Lu.

Figure 1
Figure 1. Figure 1: Unsupervised object detection and instance segmentation. The single-frame comparison (left) shows a comparison between our annotation module VitCut (used in the preprocessing stage of VVitCutLER) and the reference method VoteCut. VitCut generates significantly higher quality pseudomasks. The video example (right) is generated by the complete VVitCutLER system and compared with the state-of-the-art unsuperv… view at source ↗
Figure 2
Figure 2. Figure 2: The upper part illustrates the complete two-step process of [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of VVitCutLER. Unlabeled images are pro￾cessed by VitCut to produce pseudo masks and boxes, which are used as pseudo labels to train the detector. We adopt self-training, reusing current predictions as pseudo labels for the next round. Within the detector, a SELSA module follows the box head to ag￾gregate temporal features for video learning. As shown in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training loss comparison between bbox-only aggre [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Overview of the proposed VideoCut framework for unsupervised mask extraction. Step 1: Multiple ViT model, NCut, and [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: CutSAM’s overall architecture. The detected bounding boxes are further clustered to group related regions, and SAM2 is then [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative visualizations on YouTube-VIS 2021, DAVIS, and ImageNet-VID. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative visualizations on YouTube-VIS 2019 and OVIS. We compare different annotation generation methods on two additional video instance segmentation benchmarks that are not used for training. The results show that VitCut produces more stable and refined masks across diverse scenarios, indicating strong cross-dataset generalization. K ∈ {30, 100, 120, 150, 200} and calculated the average recall (AR) on… view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative visualizations of VVitCutLER on YouTube-VIS 2021 and ImageNet-VID. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Effect of different TopK values on AR and runtime at IoU threshold 0.5. TopK=150 achieves the best balance be￾tween accuracy and efficiency [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Effect of inserting the aggregator module into dif￾ferent stages of Cascade R-CNN on YouTube-VIS. The baseline achieves the best performance, while inserting the aggregator at various stages results in performance degradation. single-stage or two-stage detectors are more tolerant of tem￾poral fusion and can benefit from our aggregation design. 11.3. Effect of Teacher Model Choice To explore how different … view at source ↗
read the original abstract

Unsupervised pixel-level video understanding remains challenging in real-world scenarios, where motion blur, occlusion, and fast object dynamics often cause temporal drift and flickering pseudo-labels.We propose VVitCutLER, an unsupervised framework for video object detection and instance segmentation, which improves the quality of pseudo-labels through temporal consistency. Our core contribution is VitCut, a temporarily stable pseudo-label generator that reduces error accumulation during field degradation through cross-frame region consistency. Meanwhile, VitCut uses a distillation decoder to achieve effective instance mask prediction. Then, based on VitCut, VVitCutLER further integrates cross-frame feature aggregation to enhance video-level robustness. Extensive experiments on standard video benchmarks demonstrate that VVitCutLER significantly improves detection and segmentation performance while reducing temporal instability. These results highlight the importance of temporally consistent supervision for robust pixel-level video understanding.

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 proposes VVitCutLER, an unsupervised framework for video object detection and instance segmentation. Its core contribution is VitCut, a pseudo-label generator that enforces cross-frame region consistency to produce temporally stable labels and reduce error accumulation from motion blur, occlusion, and fast dynamics. It incorporates a distillation decoder for instance mask prediction and adds cross-frame feature aggregation for video-level robustness. Experiments on standard video benchmarks are reported to demonstrate significant gains in detection/segmentation performance together with reduced temporal instability.

Significance. If the central claims are substantiated, the work could advance unsupervised pixel-level video understanding by showing how explicit temporal consistency mechanisms can mitigate drift and flickering in pseudo-labels. The emphasis on cross-frame consistency as a means to improve robustness in challenging real-world conditions is a relevant direction for the field.

major comments (2)
  1. [VitCut description] VitCut section: The claim that cross-frame region consistency reliably reduces error accumulation (rather than propagating initial pseudo-label noise) is load-bearing for the central contribution. The manuscript must provide concrete analysis or ablations showing that the matching mechanism correctly identifies corresponding regions under the occlusion and fast-motion cases explicitly listed as challenges; without this, the risk that consistency locks in or spreads errors remains unaddressed.
  2. [Experiments] Experiments section: Reported improvements in detection and segmentation must be accompanied by quantitative temporal-stability metrics (e.g., frame-to-frame mask IoU consistency or flicker scores) and direct comparisons against recent unsupervised video baselines; current claims of reduced instability rest on qualitative statements that are insufficient to support the headline result.
minor comments (2)
  1. [Abstract] Abstract: 'temporarily stable' is almost certainly a typo for 'temporally stable'.
  2. [Abstract] Abstract: The phrase 'during field degradation' is unclear; rephrase to specify whether feature, label, or another form of degradation is intended.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We have addressed each major comment below and revised the manuscript to incorporate additional analysis and quantitative evaluations where appropriate.

read point-by-point responses

  1. Referee: [VitCut description] VitCut section: The claim that cross-frame region consistency reliably reduces error accumulation (rather than propagating initial pseudo-label noise) is load-bearing for the central contribution. The manuscript must provide concrete analysis or ablations showing that the matching mechanism correctly identifies corresponding regions under the occlusion and fast-motion cases explicitly listed as challenges; without this, the risk that consistency locks in or spreads errors remains unaddressed.

    Authors: We appreciate the referee's emphasis on this critical aspect of our contribution. The cross-frame region consistency in VitCut is intended to enforce temporal coherence and thereby limit drift from motion blur, occlusion, and fast dynamics. To directly address the concern regarding potential error propagation, we have added new ablation studies and visualizations in the revised manuscript. These include quantitative matching accuracy metrics on challenging subsequences exhibiting occlusion and rapid motion, as well as qualitative examples demonstrating correct region correspondence. We also discuss cases where initial pseudo-label noise may be reinforced and how the overall framework mitigates this through the distillation decoder. revision: yes

  2. Referee: [Experiments] Experiments section: Reported improvements in detection and segmentation must be accompanied by quantitative temporal-stability metrics (e.g., frame-to-frame mask IoU consistency or flicker scores) and direct comparisons against recent unsupervised video baselines; current claims of reduced instability rest on qualitative statements that are insufficient to support the headline result.

    Authors: We agree that explicit quantitative metrics are necessary to substantiate claims of reduced temporal instability. In the revised manuscript, we now report frame-to-frame mask IoU consistency and a flicker score computed across video sequences on the evaluated benchmarks. We have also included direct comparisons against recent unsupervised video object detection and segmentation baselines. These additions provide empirical support for improved stability alongside the reported gains in detection and segmentation performance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on novel architectural components and empirical evaluation

full rationale

The paper introduces VitCut as a new pseudo-label generator enforcing cross-frame region consistency plus a distillation decoder, then builds VVitCutLER by adding cross-frame feature aggregation. These are presented as independent design choices whose value is assessed via experiments on standard video benchmarks. No equations, fitted parameters, or self-citations are shown to reduce the central performance claims to tautological redefinitions or inputs by construction. The framework is self-contained against external benchmarks and does not invoke uniqueness theorems or prior self-work as load-bearing justification.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is provided, so no concrete free parameters, axioms, or invented entities can be extracted or audited.

pith-pipeline@v0.9.0 · 5681 in / 1034 out tokens · 30749 ms · 2026-05-20T22:26:10.144196+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · 1 internal anchor

  1. [1]

    Cuvler: Enhanced unsupervised object discoveries through exhaustive self-supervised transformers, 2024

    Shahaf Arica, Or Rubin, Sapir Gershov, and Shlomi Laufer. Cuvler: Enhanced unsupervised object discoveries through exhaustive self-supervised transformers, 2024. 2, 3

    work page 2024

  2. [2]

    Cascade r-cnn: Delving into high quality object detection, 2017

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: Delving into high quality object detection, 2017. 6

    work page 2017

  3. [3]

    End-to- end object detection with transformers, 2020

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to- end object detection with transformers, 2020. 2

    work page 2020

  4. [4]

    Emerg- ing properties in self-supervised vision transformers, 2021

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers, 2021. 3

    work page 2021

  5. [5]

    Tclr: Temporal contrastive learning for video representation.Computer Vision and Image Understanding, 219:103406, 2022

    Ishan Dave, Rohit Gupta, Mamshad Nayeem Rizve, and Mubarak Shah. Tclr: Temporal contrastive learning for video representation.Computer Vision and Image Understanding, 219:103406, 2022. 3

    work page 2022

  6. [6]

    Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023

    Henghui Ding, Chang Liu, Shuting He, Xudong Jiang, and Chen Change Loy. Mevis: A large-scale benchmark for video segmentation with motion expressions, 2023. 3

    work page 2023

  7. [7]

    Henghui Ding, Chang Liu, Shuting He, Xudong Jiang, Philip H. S. Torr, and Song Bai. Mose: A new dataset for video object segmentation in complex scenes, 2023. 3

    work page 2023

  8. [8]

    Mevis: A multi-modal dataset for referring motion expres- sion video segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(12):11400–11416,

    Henghui Ding, Chang Liu, Shuting He, Kaining Ying, Xudong Jiang, Chen Change Loy, and Yu-Gang Jiang. Mevis: A multi-modal dataset for referring motion expres- sion video segmentation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(12):11400–11416,

    work page

  9. [9]

    Henghui Ding, Kaining Ying, Chang Liu, Shuting He, Xudong Jiang, Yu-Gang Jiang, Philip H. S. Torr, and Song Bai. Mosev2: A more challenging dataset for video object segmentation in complex scenes, 2025. 3

    work page 2025

  10. [10]

    Two-frame motion estimation based on polynomial expansion

    Gunnar Farneb ¨ack. Two-frame motion estimation based on polynomial expansion. InScandinavian conference on Im- age analysis, pages 363–370. Springer, 2003. 4

    work page 2003

  11. [11]

    Distributed in- telligent video surveillance for early armed robbery detection based on deep learning, 2024

    Sergio Fernandez-Testa and Edwin Salcedo. Distributed in- telligent video surveillance for early armed robbery detection based on deep learning, 2024. 1

    work page 2024

  12. [12]

    Rich feature hierarchies for accurate object detection and semantic segmentation, 2014

    Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation, 2014. 2

    work page 2014

  13. [13]

    Tokencut: Segmentation-free object discovery by clustering tokens

    Jiawen Guo, Luming Xie, Zhe Lin, and Chen Change Loy. Tokencut: Segmentation-free object discovery by clustering tokens. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 3

    work page 2022

  14. [14]

    Hadi Hadizadeh and Ivan V . Baji´c. Learned multimodal com- pression for autonomous driving, 2024. 1

    work page 2024

  15. [15]

    Unsupervised ob- ject segmentation in video by efficient selection of highly probable positive features, 2017

    Emanuela Haller and Marius Leordeanu. Unsupervised ob- ject segmentation in video by efficient selection of highly probable positive features, 2017. 3

    work page 2017

  16. [16]

    Seq-nms for video object detection

    Wei Han, Anna Khoreva, Eddy Ilg, Deqing Sun, Varun Jampani, Edward Adelson, Michael Black, Andreas Geiger, Alexey Dosovitskiy, and Thomas Brox. Seq-nms for video object detection. InICCV, 2016. 3

    work page 2016

  17. [17]

    Mask r-cnn, 2018

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn, 2018. 2

    work page 2018

  18. [18]

    Momentum contrast for unsupervised visual rep- resentation learning, 2020

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning, 2020. 3

    work page 2020

  19. [19]

    Transvod: End-to-end video object detection with spatial- temporal transformers

    Tao He, Ziyi Wu, Enze Xie, Ding Liang, and Chunhua Shen. Transvod: End-to-end video object detection with spatial- temporal transformers. InCVPR, 2022. 3

    work page 2022

  20. [20]

    Uvis: Unsupervised video instance segmentation, 2024

    Shuaiyi Huang, Saksham Suri, Kamal Gupta, Sai Saketh Rambhatla, Ser nam Lim, and Abhinav Shrivastava. Uvis: Unsupervised video instance segmentation, 2024. 7

    work page 2024

  21. [21]

    T-cnn: Tubelets with convolutional neural networks for object detection from videos

    Kai Kang, Wanli Ouyang, Hongsheng Li, and Xiaogang Wang. T-cnn: Tubelets with convolutional neural networks for object detection from videos. InICCV, 2017. 3

    work page 2017

  22. [22]

    Mo- tion guided attention for video salient object detection, 2019

    Haofeng Li, Guanqi Chen, Guanbin Li, and Yizhou Yu. Mo- tion guided attention for video salient object detection, 2019. 2

    work page 2019

  23. [23]

    Focal loss for dense object detection, 2018

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ar. Focal loss for dense object detection, 2018. 2

    work page 2018

  24. [24]

    Berg.SSD: Single Shot MultiBox Detector, page 21–37

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C. Berg.SSD: Single Shot MultiBox Detector, page 21–37. Springer International Publishing, 2016. 2

    work page 2016

  25. [25]

    Learning temporal cues by predicting ob- jects move for multi-camera 3d object detection, 2024

    Seokha Moon, Hongbeen Park, Jungphil Kwon, Jaekoo Lee, and Jinkyu Kim. Learning temporal cues by predicting ob- jects move for multi-camera 3d object detection, 2024. 1

    work page 2024

  26. [26]

    Nguyen, Tuan N

    Thuy C. Nguyen, Tuan N. Tang, Nam LH. Phan, Chuong H. Nguyen, Masayuki Yamazaki, and Masao Yamanaka. 1st place solution for youtubevos challenge 2021:video instance segmentation, 2021. 5

    work page 2021

  27. [27]

    Tubetk: Adopting tubes to track multi-object in a one-step training model, 2020

    Bo Pang, Yizhuo Li, Yifan Zhang, Muchen Li, and Cewu Lu. Tubetk: Adopting tubes to track multi-object in a one-step training model, 2020. 2

    work page 2020

  28. [28]

    The 2017 DAVIS Challenge on Video Object Segmentation

    Jordi Pont-Tuset, Federico Perazzi, Sergi Caelles, Pablo Ar- bel´aez, Alexander Sorkine-Hornung, and Luc Van Gool. The 2017 davis challenge on video object segmentation. arXiv:1704.00675, 2017. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  29. [29]

    Sam 2: Segment anything in images and videos,

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross Girshick, Piotr Doll´ar, and Christoph Feicht- enhofer. Sam 2: Segment anything in images and videos,

    work page

  30. [30]

    You only look once: Unified, real-time object de- tection, 2016

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection, 2016. 2

    work page 2016

  31. [31]

    Faster r-cnn: Towards real-time object detection with region proposal networks, 2016

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks, 2016. 2

    work page 2016

  32. [32]

    Berg, and Li Fei-Fei

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. Imagenet large scale visual recognition challenge,

    work page

  33. [33]

    Flowcut: Unsupervised video instance segmentation via temporal mask matching,

    Alp Eren Sari and Paolo Favaro. Flowcut: Unsupervised video instance segmentation via temporal mask matching,

    work page

  34. [34]

    Yolov: Making still image object detectors great at video object detection,

    Yuheng Shi, Naiyan Wang, and Xiaojie Guo. Yolov: Making still image object detectors great at video object detection,

    work page

  35. [35]

    Practical video object detection via feature selection and aggregation, 2024

    Yuheng Shi, Tong Zhang, and Xiaojie Guo. Practical video object detection via feature selection and aggregation, 2024. 3

    work page 2024

  36. [36]

    V o, Simon Roburin, Spyros Gidaris, Andrei Bursuc, Patrick P ´erez, Renaud Mar- let, and Jean Ponce

    Oriane Sim ´eoni, Gilles Puy, Huy V . V o, Simon Roburin, Spyros Gidaris, Andrei Bursuc, Patrick P ´erez, Renaud Mar- let, and Jean Ponce. Localizing objects with self-supervised transformers and no labels, 2021. 3

    work page 2021

  37. [37]

    Raft: Recurrent all-pairs field transforms for optical flow, 2020

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow, 2020. 4, 1

    work page 2020

  38. [38]

    Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training, 2022

    Zhan Tong, Yibing Song, Jue Wang, and Limin Wang. Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training, 2022. 3

    work page 2022

  39. [39]

    Yu, and Ishan Misra

    Xudong Wang, Rohit Girdhar, Stella X. Yu, and Ishan Misra. Cut and learn for unsupervised object detection and instance segmentation, 2023. 2, 3, 5

    work page 2023

  40. [40]

    Videocutler: Surprisingly simple unsuper- vised video instance segmentation, 2023

    Xudong Wang, Ishan Misra, Ziyun Zeng, Rohit Girdhar, and Trevor Darrell. Videocutler: Surprisingly simple unsuper- vised video instance segmentation, 2023. 3, 7

    work page 2023

  41. [41]

    Simple online and realtime tracking with a deep association metric,

    Nicolai Wojke, Alex Bewley, and Dietrich Paulus. Simple online and realtime tracking with a deep association metric,

    work page

  42. [42]

    Sequence level semantics aggregation for video ob- ject detection, 2019

    Haiping Wu, Yuntao Chen, Naiyan Wang, and Zhaoxiang Zhang. Sequence level semantics aggregation for video ob- ject detection, 2019. 2, 3, 5

    work page 2019

  43. [43]

    Segmenting moving objects via an object-centric layered representation,

    Junyu Xie, Weidi Xie, and Andrew Zisserman. Segmenting moving objects via an object-centric layered representation,

    work page

  44. [44]

    Liang Yan, Qing Wang, Song Ma, Jian Wang, and Chong Yu. Solve the puzzle of instance segmentation in videos: A weakly supervised framework with spatio-temporal collabo- ration.IEEE Transactions on Circuits and Systems for Video Technology, 33(1):393–406, 2023. 3

    work page 2023

  45. [45]

    Self-supervised video object segmentation by motion grouping

    Charig Yang, Hala Lamdouar, Erika Lu, Andrew Zisserman, and Weidi Xie. Self-supervised video object segmentation by motion grouping. InICCV, 2021. 7

    work page 2021

  46. [46]

    Unsupervised moving object detection via contextual information separation

    Yanchao Yang, Antonio Loquercio, Davide Scaramuzza, and Stefano Soatto. Unsupervised moving object detection via contextual information separation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 879–888, 2019. 3

    work page 2019

  47. [47]

    Bytetrack: Multi-object tracking by associating every detection box, 2022

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi-object tracking by associating every detection box, 2022. 3

    work page 2022

  48. [48]

    Flow-guided feature aggregation for video object de- tection, 2017

    Xizhou Zhu, Yujie Wang, Jifeng Dai, Lu Yuan, and Yichen Wei. Flow-guided feature aggregation for video object de- tection, 2017. 2, 3

    work page 2017

  49. [49]

    Deformable detr: Deformable transformers for end-to-end object detection, 2021

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection, 2021. 2

    work page 2021