Pith. sign in

REVIEW 3 major objections 6 minor 87 references

Learning Motion and Temporal Cues for Unsupervised Video Object Segmentation

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read MTNet fuses motion and temporal cues to reach state of the art in unsupervised video object segmentation.

desk verdict MTNet is a competent real-time UVOS system with plausible SOTA numbers, but the inference protocol's silence about discarded tail frames puts the main comparison on shaky ground. read the letter →

arxiv 2501.07806 v1 pith:FDVDUD73 submitted 2025-01-14 cs.CV

Yunzhi Zhuge , Hongyu Gu , Lu Zhang , Jinqing Qi , Huchuan Lu This is my paper · ORCID
classification cs.CV
keywords unsupervisedvideoobjectsegmentationsalientdetectionopticalflowtemporaltransformermulti-modalfusioncascadeddecoderlong-rangemodeling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that unsupervised video object segmentation improves when motion and temporal cues are combined in one unified framework, rather than treated separately. The proposed MTNet merges appearance and optical-flow features inside the encoder, applies a two-layer temporal transformer to model inter-frame relations, and refines masks through cascaded transformer decoders. The authors report that MTNet reaches a J&F mean of 89.7 on DAVIS-16, surpassing the previous state of the art, and is also competitive on video salient object detection while running at 43.4 fps. The paper argues that each architectural component contributes a meaningful gain and that the method is robust to corruptions and long videos.

What carries the argument

The central object is the MTNet architecture, whose three load-bearing components are the Bi-modal Fusion Module (BFM), the Mixed Temporal Transformer (MTT), and the Cascaded Transformer Decoder (CTD). BFM combines appearance and motion features inside each encoder stage using a multi-modal gate and co-attention; MTT stacks a local-window temporal transformer layer with a global temporal transformer layer that compresses keys and values to keep self-attention tractable; CTD progressively refines multi-level feature maps from deep to shallow. Together they carry the argument that motion cues and temporal context must be fused jointly, not sequentially, to track the primary object under occlusion and fast motion.

What would settle it

A direct falsifier would be an independent re-evaluation of MTNet and GSA on DAVIS-16 under identical inference settings (same clip length, same flow, no post-processing), checking whether the 0.7-point J&F gap persists; if it vanishes under a slightly different clip length or augmentation, the state-of-the-art claim would be protocol-dependent rather than architectural.

Watch

Extended reading notes

Core claim

The central claim is that a single end-to-end network can exploit both cross-modal (RGB plus optical flow) and cross-frame (temporal) information more effectively than prior methods that use one or the other. Concretely, MTNet's Bi-modal Fusion Module fuses appearance and motion features at every encoder stage using co-channel and co-spatial attention, while the Mixed Temporal Transformer combines a local-window temporal self-attention layer with a global summarization layer to capture long-range temporal dependencies. The Cascaded Transformer Decoder then propagates information from deep to shallow levels. On DAVIS-16 the method reports 89.7 J&F mean, beating GSA's 89.0, and on FBMS it reports 83.8 J mean, beating TMO's 79.9, with the ablation study attributing the gain to the joint use of all three components rather than to any single module.

Load-bearing premise

The reported state-of-the-art margins rely on the assumption that the comparison protocol is fair and stable: the re-measured baseline numbers and the shared training recipe (YouTube-VOS pretraining, DAVIS-16 fine-tuning, three-frame sampling) match those of the prior methods they are compared with.

Editorial extensions

If this is right

  • MTNet's reported numbers on DAVIS-16, YouTube-Objects, FBMS, and Long-Videos suggest that joint motion-temporal fusion can replace heavier post-processing such as CRF or multi-scale testing without sacrificing accuracy.
  • The method's clip-level inference at 43.4 fps on a 2080Ti indicates that a transformer-based temporal model can be practical for near-real-time UVOS, not just offline processing.
  • The same network, without dataset-specific fine-tuning, transfers to video salient object detection on ViSal, SegTrack-V2, and DAVSOD, implying that the learned objectness and motion priors are task-generic.
  • The ablation results imply that each of the three proposed modules contributes positively, with the full combination needed to reach the reported state-of-the-art numbers.

Reading between the lines

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

  • The role of the Mixed Temporal Transformer could be tested more sharply by comparing against a plain 3D conv or a single non-local block with similar parameter count; the paper's ablation isolates modules but not the temporal attention mechanism itself.
  • Because the method clips videos into fixed-length segments, the improvement on Long-Videos as clip length grows suggests a saturation point that, if mapped precisely, could tell whether longer-range memory beyond one clip would help or whether per-clip processing is enough.
  • The robustness to ImageNet-C corruptions is tested only on DAVIS-16; extending the corruption protocol to FBMS and Long-Videos would clarify whether the reported robustness is specific to the benchmark's object scale and motion patterns.
  • The paper's failure cases show confusion in multi-instance and crowded scenes, so a natural extension would be to add an explicit instance-discrimination or slot-based mechanism, which recent slot-attention UVOS methods already explore.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. MTNet is a two-stream unsupervised video object segmentation architecture that fuses RGB and RAFT optical flow through a Bi-modal Fusion Module, models cross-frame dependencies with a Mixed Temporal Transformer, and refines multi-scale features via Cascaded Transformer Decoders. The paper reports state-of-the-art results on DAVIS-16 (J&F 89.7), FBMS (J mean 83.8), YouTube-Objects (J mean 74.6), and Long-Videos, plus competitive results on four VSOD benchmarks, with inference speed of 43.4 fps. The manuscript includes extensive ablations, robustness tests, visualizations, a failure-case analysis, and a link to source code.

Significance. Assuming the evaluation protocol is fair and fully specified, this is a solid empirical contribution. The architecture is a coherent combination of known components, the ablations in Table VI show that each proposed module contributes, and the robustness study in Fig. 7 and the failure analysis in Fig. 9 add useful evidence. The main risk is not the architecture itself but the measurement protocol: the inference rule C = floor(N/T) defined in Section IV-A.5 can discard trailing frames, which would make the headline comparisons with methods evaluated on full benchmark sequences unreliable. The paper also makes source code available, which is a reproducibility strength.

major comments (3)
  1. [IV-A.5 / Tables I and III] The inference protocol partitions a test video of N frames into C = floor(N/T) clips and obtains clip-level results directly, leaving N mod T frames unprocessed. The paper never states how these tail frames enter the reported J&F means. For example, a 30-frame DAVIS-16 sequence with T = 12 would have 6 of 30 frames excluded if the text is taken literally, so the reported 89.7 J&F in Table I would not be computed over the same frames as the 89.0 reported for GSA. The FBMS result in Table III is subject to the same ambiguity. Please specify the exact tail-frame handling (padding, overlapping clips, a final shorter clip, or deliberate exclusion), state the number of frames actually scored per benchmark, and provide results evaluated over all annotated frames.
  2. [IV-B / Table I] The central state-of-the-art claims rest on margins of 0.7 J&F on DAVIS-16 and 3.9 J mean on FBMS, but no error bars, confidence intervals, or number of training runs are reported. Because training uses random augmentations, stochastic clip sampling, and random initialization, a single run may not reliably establish superiority over GSA or TMO. Please report mean and standard deviation over at least three independent runs, or clearly justify why single-run benchmark numbers are sufficient under the evaluation protocol used.
  3. [IV-B.5 / Table V] The re-measured baseline numbers marked with asterisks in Table V are load-bearing for the VSOD comparisons, but the text gives no details about the re-measurement protocol, including preprocessing, clip length, frame coverage, and whether the same floor-based clip partitioning was used. Without these details, the reader cannot judge whether the improvements over HFAN and other methods are due to the architecture or to protocol differences. Please document the re-measurement settings, release the evaluation scripts, and clarify which entries were re-measured and under which exact conditions.
minor comments (6)
  1. [III-B, Eq. (1)] The text says the appearance and motion features are initially compressed with two separate 3x3 convolutional layers, but Eq. (1) and the subsequent definition of ConvS describe a 1x1 convolution; please reconcile the notation.
  2. [III-C] The local window dimension is described as 'T x W x W x d'; this should presumably read 'T x M x M x d' to match the stated windowing into H*W/M^2 windows.
  3. [Table IV] The citation labels in Table IV do not match the bibliography: STM is reference [4] rather than [15], AFB-URR is reference [33] rather than [85], AOT is reference [73] rather than [82], and HFAN is reference [47] rather than [82]. Please renumber the table entries.
  4. [Table I] In the DFNet row of Table I, the F Mean value '8.8' appears to be a typo, likely '88.0'; please verify all numeric entries for consistency with the metric definitions.
  5. [Conclusion / III-C] The module is called 'Mixed Temporal Transformer' in Section III-C but 'Mixed Temporal Module' in the conclusion; please use a single consistent name throughout.
  6. [Abstract and code link] Providing a commit hash or an archived version of the repository would improve reproducibility, since the current link points to an unversioned repository state.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MTNet's performance claims come from benchmark evaluation of an independently trained architecture, not from a derivation that assumes its conclusions.

full rationale

The paper's central claims are empirical: MTNet is trained with a BCE loss against ground-truth masks on YouTube-VOS and DAVIS-16 and then evaluated on held-out benchmarks (DAVIS-16 validation, FBMS test, YouTube-Objects, Long-Videos). No stated equation or module definition reduces a reported number to a fitted input. Hyperparameters such as lambda = 0.5 and test clip length t = 12 are set by hand or by ablation, and even if they were tuned on validation data, that would be standard model selection rather than circularity. The VSOD table's re-measured baselines are obtained 'employing the original testing codes and model weights in their projects' (Section IV-B.5), which is external evidence, not self-citation. The paper does not invoke a uniqueness theorem from the authors' own prior work, and although it follows HFAN for training details and RAFT for optical flow, those are external methods cited for procedure, not for the target result. The reviewer-flagged concern about clip partitioning (C = floor(N/T), Section IV-A.5, potentially discarding tail frames) is a possible evaluation-protocol fairness issue affecting comparability with baselines, but it is not circularity: the output masks are still produced by the trained network and scored by the official metric. No circular step can be exhibited, so the score is 0.

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

The paper introduces no physical entities and no fitted physical constants. Its free parameters are standard training hyperparameters. The main unverified inputs are pretrained weights (ConvNeXt, RAFT) and benchmark labels.

free parameters (4)
  • loss weight lambda = 0.5
    Set to balance auxiliary BCE losses in Eq. (25); chosen by hand, not tuned per benchmark.
  • training clip length T = 3
    Three frames sampled per clip during training, following STCN; affects temporal modeling.
  • inference clip length t = 12
    Selected based on Long-Videos saturation in Fig. 6; affects performance on long videos.
  • encoder feature stages K = 4
    ConvNeXt has four stages; BFM is applied at all stages and MTT only at stages 3 and 4.
assumptions (4)
  • domain assumption Pretrained ConvNeXt weights and RAFT optical flow provide reliable feature extractors without task-specific adaptation.
    The method builds on these pretrained models; if flow is noisy or backbone features are weak, performance degrades, as the motion-only ablation shows a large drop.
  • domain assumption Benchmark ground-truth masks are accurate and the evaluation metrics are uniformly applied.
    All state-of-the-art comparisons assume the datasets and metrics are used consistently across methods, including re-measured baselines.
  • standard math Standard transformer attention, Softmax(QK^T / sqrt(d)) V, is a valid building block.
    Used in MTT and CTD without proof; accepted in the literature.
  • domain assumption Pretraining on YouTube-VOS followed by fine-tuning on DAVIS-16 is a valid transfer setting that does not unfairly favor MTNet.
    The authors follow HFAN's protocol; if prior methods used less data, comparisons would be unfair.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Motion and Temporal Cues for Unsupervised Video Object Segmentation." pith.science (2026). https://pith.science/paper/FDVDUD73

@misc{pith2026250107806,
  author       = {Pith},
  title        = {Pith review of: Learning Motion and Temporal Cues for Unsupervised Video Object Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FDVDUD73}},
  note         = {Machine review of arXiv:2501.07806}
}
read the original abstract

In this paper, we address the challenges in unsupervised video object segmentation (UVOS) by proposing an efficient algorithm, termed MTNet, which concurrently exploits motion and temporal cues. Unlike previous methods that focus solely on integrating appearance with motion or on modeling temporal relations, our method combines both aspects by integrating them within a unified framework. MTNet is devised by effectively merging appearance and motion features during the feature extraction process within encoders, promoting a more complementary representation. To capture the intricate long-range contextual dynamics and information embedded within videos, a temporal transformer module is introduced, facilitating efficacious inter-frame interactions throughout a video clip. Furthermore, we employ a cascade of decoders all feature levels across all feature levels to optimally exploit the derived features, aiming to generate increasingly precise segmentation masks. As a result, MTNet provides a strong and compact framework that explores both temporal and cross-modality knowledge to robustly localize and track the primary object accurately in various challenging scenarios efficiently. Extensive experiments across diverse benchmarks conclusively show that our method not only attains state-of-the-art performance in unsupervised video object segmentation but also delivers competitive results in video salient object detection. These findings highlight the method's robust versatility and its adeptness in adapting to a range of segmentation tasks. Source code is available on https://github.com/hy0523/MTNet.

Figures

Figures reproduced from arXiv: 2501.07806 by the authors.

Figure 1
Figure 1. Three distinct methodologies for unsupervised video [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) The proposed MTNet pipeline utilizes [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of (a) Local Window MHSA and (b) Global [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Visual analysis of unsupervised video object segmentation performance across a variety of video scenarios. The sequence [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Visual comparison of the saliency maps between our methods and state-of-the-art models. This figure illustrates the [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Impact of Clip Length on J &F Mean Score Across Datasets. The line chart shows an improvement trend with longer clips in Long-Videos, contrasting with the negligible changes in DAVIS-16 and FBMS, potentially due to their less complex scenarios, which typically feature …
Figure 8
Figure 8. Figure 8: Visualizations of various components in our pro [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Illustrative failures of MTNet in Unsupervised Video [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

87 extracted references · 72 canonical work pages

  1. [85]

    Motion-attentive transition for zero- shot video object segmentation

    Tianfei Zhou et al. “Motion-attentive transition for zero- shot video object segmentation”. In: Proceedings of the AAAI conference on artificial intelligence . 2020, pp. 13066–13073

    work page 2020

  2. [1]

    Frequency-tuned salient region detection

    Radhakrishna Achanta et al. “Frequency-tuned salient region detection”. In: IEEE Conf. Comput. Vis. Pattern Recog. IEEE. 2009, pp. 1597–1604

    2009

  3. [2]

    Is space-time attention all you need for video under- standing?

    Gedas Bertasius, Heng Wang, and Lorenzo Torresani. “Is space-time attention all you need for video under- standing?” In: Int. Conf. Mach. Learn. V ol. 2. 3. 2021, p. 4

    2021

  4. [3]

    The 2019 davis challenge on vos: Unsupervised multi-object segmentation

    Sergi Caelles et al. “The 2019 davis challenge on vos: Unsupervised multi-object segmentation”. In:arXiv preprint arXiv:1905.00737 (2019)

    arXiv 2019

  5. [4]

    Re- thinking space-time networks with improved memory coverage for efficient video object segmentation

    Ho Kei Cheng, Yu-Wing Tai, and Chi-Keung Tang. “Re- thinking space-time networks with improved memory coverage for efficient video object segmentation”. In: Adv. Neural Inform. Process. Syst. (2021), pp. 11781– 11794

    2021

  6. [5]

    Structure- measure: A new way to evaluate foreground maps

    Ming-Ming Cheng and Deng-Ping Fan. “Structure- measure: A new way to evaluate foreground maps”. In: Int. J. Comput. Vis. (2021), pp. 2622–2638

    2021

  7. [6]

    Dual Prototype Attention for Unsupervised Video Object Segmentation

    Suhwan Cho et al. “Domain Alignment and Temporal Aggregation for Unsupervised Video Object Segmenta- tion”. In: arXiv preprint arXiv:2211.12036 (2022)

    work page Pith review arXiv 2022

  8. [7]

    Treating motion as option to re- duce motion dependency in unsupervised video object segmentation

    Suhwan Cho et al. “Treating motion as option to re- duce motion dependency in unsupervised video object segmentation”. In: IEEE Int. Conf. Appli. Comp. Vision. 2023, pp. 5140–5149

    2023

Show all 87 references
  1. [8]

    High-performance long-term track- ing with meta-updater

    Kenan Dai et al. “High-performance long-term track- ing with meta-updater”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2020, pp. 6298–6307

    2020

  2. [9]

    FEANet: Feature-enhanced attention network for RGB-thermal real-time semantic segmen- tation

    Fuqin Deng et al. “FEANet: Feature-enhanced attention network for RGB-thermal real-time semantic segmen- tation”. In: Int. Conf. Intel. Robots Syst. IEEE. 2021, pp. 4467–4473

    2021

  3. [10]

    Scaling up your kernels to 31x31: Revisiting large kernel design in cnns

    Xiaohan Ding et al. “Scaling up your kernels to 31x31: Revisiting large kernel design in cnns”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2022, pp. 11963–11975

    2022

  4. [11]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy et al. “An image is worth 16x16 words: Transformers for image recognition at scale”. In: Int. Conf. Learn. Represent. 2020

    2020

  5. [12]

    Enhanced-alignment measure for binary foreground map evaluation

    Deng-Ping Fan et al. “Enhanced-alignment measure for binary foreground map evaluation”. In: arXiv preprint arXiv:1805.10421 (2018)

    2018 arXiv

  6. [13]

    Shifting more attention to video salient object detection

    Deng-Ping Fan et al. “Shifting more attention to video salient object detection”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2019, pp. 8554–8564

    2019

  7. [14]

    Multiscale vision transformers

    Haoqi Fan et al. “Multiscale vision transformers”. In: Int. Conf. Comput. Vis. 2021, pp. 6824–6835

    2021

  8. [15]

    Pyramid constrained self-attention network for fast video salient object detection

    Yuchao Gu et al. “Pyramid constrained self-attention network for fast video salient object detection”. In: AAAI Conf. Arti. Intell. 2020, pp. 10869–10876

    2020

  9. [16]

    Cmt: Convolutional neural net- works meet vision transformers

    Jianyuan Guo et al. “Cmt: Convolutional neural net- works meet vision transformers”. In: IEEE Conf. Com- put. Vis. Pattern Recog. 2022, pp. 12175–12185

    2022

  10. [17]

    Benchmarking neural network robustness to common corruptions and perturbations

    Dan Hendrycks and Thomas Dietterich. “Benchmarking neural network robustness to common corruptions and perturbations”. In: Int. Conf. Learn. Represent. 2019

    2019

  11. [18]

    Squeeze-and- excitation networks

    Jie Hu, Li Shen, and Gang Sun. “Squeeze-and- excitation networks”. In: IEEE Conf. Comput. Vis. Pat- tern Recog. 2018, pp. 7132–7141

    2018

  12. [19]

    Goal-oriented Autonomous Driving

    Yihan Hu et al. “Goal-oriented Autonomous Driving”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2023

    2023

  13. [20]

    Unsupervised video object segmentation us- ing motion saliency-guided spatio-temporal propaga- tion

    Yuan-Ting Hu, Jia-Bin Huang, and Alexander G Schwing. “Unsupervised video object segmentation us- ing motion saliency-guided spatio-temporal propaga- tion”. In: Eur. Conf. Comput. Vis. 2018, pp. 786–802

    2018

  14. [21]

    Video instance segmentation us- ing inter-frame communication transformers

    Sukjun Hwang et al. “Video instance segmentation us- ing inter-frame communication transformers”. In: Adv. Neural Inform. Process. Syst. (2021), pp. 13352–13363

    2021

  15. [22]

    Full-duplex strategy for video ob- ject segmentation

    Ge-Peng Ji et al. “Full-duplex strategy for video ob- ject segmentation”. In: Int. Conf. Comput. Vis. 2021, pp. 4922–4933

    2021

  16. [23]

    CASNet: A cross-attention siamese net- work for video salient object detection

    Yuzhu Ji et al. “CASNet: A cross-attention siamese net- work for video salient object detection”. In: IEEE Trans. Neural Netw. Learn. Syst. 32.6 (2020), pp. 2676–2690. DOI: 10.1109/TNNLS.2020.3007534

    2020

  17. [24]

    Efficient in- ference in fully connected crfs with gaussian edge po- tentials

    Philipp Kr ¨ahenb¨uhl and Vladlen Koltun. “Efficient in- ference in fully connected crfs with gaussian edge po- tentials”. In: Adv. Neural Inform. Process. Syst. (2011)

    2011

  18. [25]

    Guided Slot Attention for Unsu- pervised Video Object Segmentation

    Minhyeok Lee et al. “Guided Slot Attention for Unsu- pervised Video Object Segmentation”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2024

    2024

  19. [26]

    Unsupervised Video Object Seg- mentation via Prototype Memory Network

    Minhyeok Lee et al. “Unsupervised Video Object Seg- mentation via Prototype Memory Network”. In: IEEE Int. Conf. Appli. Comp. Vision . 2023, pp. 5924–5934

    2023

  20. [27]

    Tsanet: Temporal and Scale Alignment for Unsupervised Video Object Segmenta- tion

    Seunghoon Lee et al. “Tsanet: Temporal and Scale Alignment for Unsupervised Video Object Segmenta- tion”. In: 2023 IEEE International Conference on Image Processing (ICIP). IEEE. 2023, pp. 1535–1539

    2023

  21. [28]

    Itera- tively selecting an easy reference frame makes unsu- pervised video object segmentation easier

    Youngjo Lee, Hongje Seong, and Euntai Kim. “Itera- tively selecting an easy reference frame makes unsu- pervised video object segmentation easier”. In: AAAI Conf. Arti. Intell. 2022, pp. 1245–1253

    2022

  22. [29]

    Video segmentation by tracking many figure-ground segments

    Fuxin Li et al. “Video segmentation by tracking many figure-ground segments”. In: Int. Conf. Comput. Vis. 2013, pp. 2192–2199

    2013

  23. [30]

    Self Supervised Progressive Network for High Performance Video Object Segmentation

    Guorong Li et al. “Self Supervised Progressive Network for High Performance Video Object Segmentation”. In: IEEE Trans. Neural Netw. Learn. Syst. (2022). 14

    2022

  24. [31]

    Efficient long-short temporal attention network for unsupervised video object segmentation

    Ping Li et al. “Efficient long-short temporal attention network for unsupervised video object segmentation”. In: Pattern Recognition 146 (2024), p. 110078

    2024

  25. [32]

    Instance embedding transfer to un- supervised video object segmentation

    Siyang Li et al. “Instance embedding transfer to un- supervised video object segmentation”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2018, pp. 6526–6535

    2018

  26. [33]

    Video object segmentation with adaptive feature bank and uncertain-region refinement

    Yongqing Liang et al. “Video object segmentation with adaptive feature bank and uncertain-region refinement”. In: Adv. Neural Inform. Process. Syst. (2020), pp. 3430– 3441

    2020

  27. [34]

    F2net: Learning to focus on the fore- ground for unsupervised video object segmentation

    Daizong Liu et al. “F2net: Learning to focus on the fore- ground for unsupervised video object segmentation”. In: AAAI Conf. Arti. Intell. 2021, pp. 2109–2117

    2021

  28. [35]

    CMX: Cross-modal fusion for RGB- X semantic segmentation with transformers

    Huayao Liu et al. “CMX: Cross-modal fusion for RGB- X semantic segmentation with transformers”. In: arXiv preprint arXiv:2203.04838 (2022)

    2022 arXiv

  29. [36]

    Learning Complementary Spatial- Temporal Transformer for Video Salient Object Detec- tion

    Nian Liu et al. “Learning Complementary Spatial- Temporal Transformer for Video Salient Object Detec- tion”. In: IEEE Trans. Neural Netw. Learn. Syst. (2023), pp. 1–11. DOI: 10.1109/TNNLS.2023.3243246

    2023

  30. [37]

    A survey of visual transformers

    Yang Liu et al. “A survey of visual transformers”. In: IEEE Trans. Neural Netw. Learn. Syst. (2023)

    2023

  31. [38]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu et al. “Swin transformer: Hierarchical vision transformer using shifted windows”. In: Int. Conf. Com- put. Vis. 2021, pp. 10012–10022

    2021

  32. [39]

    Video swin transformer

    Ze Liu et al. “Video swin transformer”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2022, pp. 3202–3211

    2022

  33. [40]

    A convnet for the 2020s

    Zhuang Liu et al. “A convnet for the 2020s”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2022, pp. 11976– 11986

    2022

  34. [41]

    Fully convolutional networks for semantic segmenta- tion

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. “Fully convolutional networks for semantic segmenta- tion”. In: IEEE Conf. Comput. Vis. Pattern Recog.2015, pp. 3431–3440

    2015

  35. [42]

    Learning video object segmenta- tion from unlabeled videos

    Xiankai Lu et al. “Learning video object segmenta- tion from unlabeled videos”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020, pp. 8960–8970

    2020

  36. [43]

    See more, know more: Unsupervised video object segmentation with co-attention siamese networks

    Xiankai Lu et al. “See more, know more: Unsupervised video object segmentation with co-attention siamese networks”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2019, pp. 3623–3632

    2019

  37. [44]

    Mixed precision training

    Paulius Micikevicius et al. “Mixed precision training”. In: arXiv preprint arXiv:1710.03740 (2017)

    2017 arXiv

  38. [45]

    Seg- mentation of moving objects by long term video analy- sis

    Peter Ochs, Jitendra Malik, and Thomas Brox. “Seg- mentation of moving objects by long term video analy- sis”. In: IEEE Trans. Pattern Anal. Mach. Intell. (2013), pp. 1187–1200

    2013

  39. [46]

    Fast object segmentation in unconstrained video

    Anestis Papazoglou and Vittorio Ferrari. “Fast object segmentation in unconstrained video”. In: Int. Conf. Comput. Vis. 2013, pp. 1777–1784

    2013

  40. [47]

    Hierarchical feature alignment net- work for unsupervised video object segmentation

    Gensheng Pei et al. “Hierarchical feature alignment net- work for unsupervised video object segmentation”. In: Eur. Conf. Comput. Vis. Springer. 2022, pp. 596–613

    2022

  41. [48]

    Hierarchical Graph Pattern Under- standing for Zero-Shot Video Object Segmentation

    Gensheng Pei et al. “Hierarchical Graph Pattern Under- standing for Zero-Shot Video Object Segmentation”. In: IEEE Trans. Image Process. 32 (2023), pp. 5909–5920. DOI: 10.1109/TIP.2023.3326395

    2023

  42. [49]

    A benchmark dataset and evaluation methodology for video object segmentation

    Federico Perazzi et al. “A benchmark dataset and evaluation methodology for video object segmentation”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2016, pp. 724–732

    2016

  43. [50]

    Saliency filters: Contrast based filtering for salient region detection

    Federico Perazzi et al. “Saliency filters: Contrast based filtering for salient region detection”. In: IEEE Conf. Comput. Vis. Pattern Recog. IEEE. 2012, pp. 733–740

    2012

  44. [51]

    Learning object class detectors from weakly annotated video

    Alessandro Prest et al. “Learning object class detectors from weakly annotated video”. In: IEEE Conf. Comput. Vis. Pattern Recog. IEEE. 2012, pp. 3282–3289

    2012

  45. [52]

    Optical flow augmented seman- tic segmentation networks for automated driving

    Hazem Rashed et al. “Optical flow augmented seman- tic segmentation networks for automated driving”. In: arXiv preprint arXiv:1901.07355 (2019)

    2019 arXiv

  46. [53]

    Reciprocal transformations for un- supervised video object segmentation

    Sucheng Ren et al. “Reciprocal transformations for un- supervised video object segmentation”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2021, pp. 15455–15464

    2021

  47. [54]

    Pyramid dilated deeper convl- stm for video salient object detection

    Hongmei Song et al. “Pyramid dilated deeper convl- stm for video salient object detection”. In: Eur. Conf. Comput. Vis. 2018, pp. 715–731

    2018

  48. [55]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. “Raft: Recurrent all-pairs field transforms for optical flow”. In: Eur. Conf. Com- put. Vis. Springer. 2020, pp. 402–419

    2020

  49. [56]

    Learning video object segmentation with visual memory

    Pavel Tokmakov, Karteek Alahari, and Cordelia Schmid. “Learning video object segmentation with visual memory”. In: Int. Conf. Comput. Vis. 2017, pp. 4481–4490

    2017

  50. [57]

    Attention is all you need

    Ashish Vaswani et al. “Attention is all you need”. In: Adv. Neural Inform. Process. Syst. (2017)

    2017

  51. [58]

    MASK-RL: Multiagent video object segmentation framework through reinforcement learning

    Giuseppe Vecchio et al. “MASK-RL: Multiagent video object segmentation framework through reinforcement learning”. In: IEEE Trans. Neural Netw. Learn. Syst. (2020), pp. 5103–5115

    2020

  52. [59]

    Con- sistent video saliency using local gradient flow opti- mization and global refinement

    Wenguan Wang, Jianbing Shen, and Ling Shao. “Con- sistent video saliency using local gradient flow opti- mization and global refinement”. In: IEEE Trans. Image Process. (2015), pp. 4185–4196

    2015

  53. [60]

    Video salient object detection via fully convolutional net- works

    Wenguan Wang, Jianbing Shen, and Ling Shao. “Video salient object detection via fully convolutional net- works”. In: IEEE Trans. Image Process. (2017), pp. 38– 49

    2017

  54. [61]

    Learning unsupervised video object segmentation through visual attention

    Wenguan Wang et al. “Learning unsupervised video object segmentation through visual attention”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2019, pp. 3064– 3074

    2019

  55. [62]

    Saliency-aware video object seg- mentation

    Wenguan Wang et al. “Saliency-aware video object seg- mentation”. In: IEEE transactions on pattern analysis and machine intelligence 40.1 (2017), pp. 20–33

    2017

  56. [63]

    Salient object detection in the deep learning era: An in-depth survey

    Wenguan Wang et al. “Salient object detection in the deep learning era: An in-depth survey”. In: IEEE Trans- actions on Pattern Analysis and Machine Intelligence 44.6 (2021), pp. 3239–3259

    2021

  57. [64]

    Zero-shot video object segmenta- tion via attentive graph neural networks

    Wenguan Wang et al. “Zero-shot video object segmenta- tion via attentive graph neural networks”. In: Int. Conf. Comput. Vis. 2019, pp. 9236–9245

    2019

  58. [65]

    Pyramid vision transformer: A versatile backbone for dense prediction without convo- lutions

    Wenhai Wang et al. “Pyramid vision transformer: A versatile backbone for dense prediction without convo- lutions”. In: Int. Conf. Comput. Vis. 2021, pp. 568–578. 15

    2021

  59. [66]

    Non-local neural networks

    Xiaolong Wang et al. “Non-local neural networks”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2018, pp. 7794–7803

    2018

  60. [67]

    Multimodal token fusion for vision transformers

    Yikai Wang et al. “Multimodal token fusion for vision transformers”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2022, pp. 12186–12195

    2022

  61. [68]

    End-to-end video instance segmen- tation with transformers

    Yuqing Wang et al. “End-to-end video instance segmen- tation with transformers”. In: IEEE Conf. Comput. Vis. Pattern Recog. 2021, pp. 8741–8750

    2021

  62. [69]

    Youtube-vos: Sequence-to-sequence video object segmentation

    Ning Xu et al. “Youtube-vos: Sequence-to-sequence video object segmentation”. In: Eur. Conf. Comput. Vis. 2018, pp. 585–601

    2018

  63. [70]

    Video enhancement with task- oriented flow

    Tianfan Xue et al. “Video enhancement with task- oriented flow”. In: Int. J. Comput. Vis.(2019), pp. 1106– 1125

    2019

  64. [71]

    Semi-supervised video salient object detection using pseudo-labels

    Pengxiang Yan et al. “Semi-supervised video salient object detection using pseudo-labels”. In: Int. Conf. Comput. Vis. 2019, pp. 7284–7293

    2019

  65. [72]

    Learning motion-appearance co- attention for zero-shot video object segmentation

    Shu Yang et al. “Learning motion-appearance co- attention for zero-shot video object segmentation”. In: Int. Conf. Comput. Vis. 2021, pp. 1564–1573

    2021

  66. [73]

    Associating objects with transformers for video object segmenta- tion

    Zongxin Yang, Yunchao Wei, and Yi Yang. “Associating objects with transformers for video object segmenta- tion”. In: Adv. Neural Inform. Process. Syst. (2021), pp. 2491–2502

    2021

  67. [74]

    Directional deep embedding and appearance learning for fast video object segmentation

    Yingjie Yin et al. “Directional deep embedding and appearance learning for fast video object segmentation”. In: IEEE Trans. Neural Netw. Learn. Syst. (2021), pp. 3884–3894

    2021

  68. [75]

    Learning joint spatial-temporal transformations for video inpainting

    Yanhong Zeng, Jianlong Fu, and Hongyang Chao. “Learning joint spatial-temporal transformations for video inpainting”. In: Eur. Conf. Comput. Vis. Springer. 2020, pp. 528–543

    2020

  69. [76]

    Adaptive semantic-enhanced trans- former for image captioning

    Jing Zhang et al. “Adaptive semantic-enhanced trans- former for image captioning”. In: IEEE Trans. Neural Netw. Learn. Syst. (2022), pp. 1–12. DOI: 10 . 1109 / TNNLS.2022.3185320

    2022

  70. [77]

    Deep transport network for un- supervised video object segmentation

    Kaihua Zhang et al. “Deep transport network for un- supervised video object segmentation”. In: Int. Conf. Comput. Vis. 2021, pp. 8781–8790

    2021

  71. [78]

    Dynamic context-sensitive filtering network for video salient object detection

    Miao Zhang et al. “Dynamic context-sensitive filtering network for video salient object detection”. In: Int. Conf. Comput. Vis. 2021, pp. 1553–1563

    2021

  72. [79]

    Learning regression and verifica- tion networks for robust long-term tracking

    Yunhua Zhang et al. “Learning regression and verifica- tion networks for robust long-term tracking”. In: Int. J. Comput. Vis. (2021), pp. 2536–2547

    2021

  73. [80]

    Mitigating modality discrepancies for RGB-T semantic segmentation

    Shenlu Zhao et al. “Mitigating modality discrepancies for RGB-T semantic segmentation”. In: IEEE Trans. Neural Netw. Learn. Syst. (2023)

    2023

  74. [81]

    Multi-source fusion and automatic predictor selection for zero-shot video object segmenta- tion

    Xiaoqi Zhao et al. “Multi-source fusion and automatic predictor selection for zero-shot video object segmenta- tion”. In: ACM Int. Conf. Multimedia . 2021, pp. 2645– 2653

    2021

  75. [82]

    Learning discriminative feature with crf for unsupervised video object segmentation

    Mingmin Zhen et al. “Learning discriminative feature with crf for unsupervised video object segmentation”. In: Eur. Conf. Comput. Vis. Springer. 2020, pp. 445– 462

    2020

  76. [83]

    Self-teaching video object seg- mentation

    Chuanwei Zhou et al. “Self-teaching video object seg- mentation”. In: IEEE Trans. Neural Netw. Learn. Syst. (2021), pp. 1623–1637

    2021

  77. [84]

    A survey on deep learning tech- nique for video segmentation

    Tianfei Zhou et al. “A survey on deep learning tech- nique for video segmentation”. In: IEEE transactions on pattern analysis and machine intelligence 45.6 (2022), pp. 7099–7122

    2022

  78. [86]

    Deep feature flow for video recogni- tion

    Xizhou Zhu et al. “Deep feature flow for video recogni- tion”. In: IEEE Conf. Comput. Vis. Pattern Recog.2017, pp. 2349–2358

    2017

  79. [87]

    Perception-aware multi- sensor fusion for 3d lidar semantic segmentation

    Zhuangwei Zhuang et al. “Perception-aware multi- sensor fusion for 3d lidar semantic segmentation”. In: Int. Conf. Comput. Vis. 2021, pp. 16280–16290

    2021

Pith tools

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