Pith. sign in

REVIEW 4 major objections 7 minor 106 references

BiVM: Accurate Binarized Neural Network for Efficient Video Matting

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read BiVM, a fully binarized video matting network, claims state-of-the-art accuracy among 1-bit matting methods — 11.82 MAD on VideoMatte240K — while cutting computation 14.3x and storage 21.6x versus its full-precision counterpart.

desk verdict The empirical gains look real, but the sparse-mask mechanism and the theory are not yet load-bearing. read the letter →

arxiv 2507.04456 v1 pith:CMRDNHMW submitted 2025-07-06 cs.CV

Haotong Qin , Xianglong Liu , Xudong Ma , Lei Ke , Yulun Zhang , Jie Luo , Michele Magno This is my paper
classification cs.CV
keywords videomattingnetworkbinarization1-bitneuralnetworksmodelcompressionefficientinferenceinformationbottleneckknowledgedistillationARMCPUdeployment
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

BiVM is built on the claim that video matting can be pushed to the extreme 1-bit regime without losing practical accuracy. The paper reports a mean absolute difference of 11.82 on the VideoMatte240K benchmark, beating 28.49 for the same matting network binarized by the generic state-of-the-art method ReActNet and 12.82 for the earlier BiMatting, and it reports 14.3x fewer FLOPs and 21.6x less storage than the full-precision counterpart, with a 13.2x latency reduction on ARM phone CPUs. The argument is that binarized matting fails for two identifiable reasons — a degenerated encoder that cannot transmit prediction-relevant information and a decoder that wastes most of its computation on homogeneous background areas — and that both can be repaired while staying fully 1-bit. If correct, real-time matting on phones and edge cameras becomes feasible at a small accuracy cost, which is the stakes of the paper.

What carries the argument

Three mechanisms carry the argument, framed by an information-bottleneck view of the network as a Markov chain Y → X → T. The Evolvable Binarized Block (EBB) is the encoder building block: non-grouped binarized convolutions with learnable layer-wise, cross-layer, and block-wise shortcuts whose scale factors are regularized to keep shortcuts from bypassing the computational units, letting a 1-bit backbone retain enough I(X;T) to produce usable features. Sparse Heterogeneous Binarization (SHB) is the decoder sparsifier: a binary incoherent-region mask computed from the 1/16-scale features by thresholding the difference between the feature map and its up-and-down-sampled versions, with the threshold chosen by maximizing the entropy of the binarized masked features, and the mask upsampled to each decoder scale to skip redundant computation. Localized Binarization-aware Mimicking (LBM) is the training mechanism: an L2 loss restricted to the mask-selected regions that aligns the binarized network's normalized multi-scale features with the full-precision counterpart's. The information-bottleneck theorems provide the theoretical glue: Theorem 1 shows the sign function cuts mutual information drastically, and Theorem 2 shows deeper transformation chains lose more information, which motivates restoring direct shortcuts.

What would settle it

Feed BiVM a probe set of videos with fine hair, translucent edges, and fast motion and check whether the alpha error concentrates inside the spatial regions where the SHB mask is zero; if it does, the mask's assumption that homogeneous regions are prediction-irrelevant is wrong. A cheaper check is to replace the entropy-maximizing threshold with a fixed mask of the same pixel density and retrain nothing — if VideoMatte240K accuracy barely moves, the entropy criterion itself, not the mask's existence, is the load-bearing part.

Watch

Extended reading notes

Core claim

The central discovery claim is that a binarized matting network's accuracy collapse is concentrated in the encoder backbone, not spread across the network, and that the efficiency bottleneck is the decoder's redundant computation on homogeneous regions; the paper then claims both are fixable at 1 bit. The fix for the encoder is the Evolvable Binarized Block (EBB), which replaces the grouped convolutions of binarized MobileNetV3 with non-grouped convolutions and threads learnable shortcuts at the layer, cross-layer, and block levels, preserving input-relevant information that the sign function would otherwise destroy (Theorem 2 formalizes the monotone loss of mutual information along a chain of transformations). The fix for the decoder is Sparse Heterogeneous Binarization (SHB), which derives a binary mask from the lowest-resolution feature map by thresholding local differences and choosing the threshold to maximize the entropy of the binarized masked features, then skips 3x3 convolutions outside the masked incoherent regions at every decoder scale. A third component, Localized Binarization-aware Mimicking (LBM), distills only the masked information-dense features from the full-precision counterpart during the first training stage. On VideoMatte240K, Distinctions-646, and Adobe Image Matting, the assembled BiVM outperforms every compared 1-bit matting network, including generic binarization methods that crash outright, and approaches the full-precision RVM's visual quality while running on ARM CPUs.

Load-bearing premise

The load-bearing premise is that a cheap binary mask, built from the smallest feature map and tuned to maximize the entropy of the binarized features it selects, reliably marks the only regions where the decoder's computation is needed — if the mask ever misses difficult regions (fine hair, translucent edges, fast motion) or the skipped regions turn out not to be redundant, both the sparsity savings and the accuracy close to full precision would weaken.

Editorial extensions

If this is right

  • A fully 1-bit network can reach 11.82 MAD on VideoMatte240K, outrunning every compared binarized matting method, including ones built on state-of-the-art generic binarization (ReActNet at 28.49 and DoReFa at 51.64, with BNN and ReCU collapsing entirely).
  • The reported 14.3x FLOP and 21.6x storage reductions against 32-bit RVM translate into a measured 13.2x latency reduction on mobile ARM CPUs, putting 512x512 matting under about 100 ms on a phone.
  • Binarized matting can beat some full-precision matting baselines: BiVM outperforms 32-bit BGMv2 on all three benchmarks and DeepLabV3 on VideoMatte240K while using a fraction of their compute.
  • Binarizing the encoder backbone is the dominant cause of accuracy collapse; fixing the backbone alone (13.46 MAD) recovers most of the gap from the 28.49 baseline, while decoder sparsity (SHB) cuts FLOPs by 52.6% once the other components are in place.

Reading between the lines

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

  • The entropy-maximizing mask criterion is a heuristic, and the paper's own ablation shows it is fragile in isolation (SHB alone collapses MAD to 191.89); a natural test is whether a mask computed from full-precision features instead of the binarized 1/16-scale features would let the sparsifier stand alone.
  • The information-bottleneck framing implies a testable design rule for other binarized dense predictors: architectures whose shortcuts raise I(X;T) without collapsing the computational path should show accuracy gains of the same order as BiVM's, which could be checked by plotting along training as in the paper's information-plane analysis.
  • Masked, information-guided distillation of the kind LBM performs could transfer to other extreme-compression settings — 2-bit quantization, mixed-precision, or post-training quantization — where uniform distillation wastes capacity on easy background regions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes BiVM, a fully binarized (1-bit) video matting network that combines an Evolvable Binarized Block (EBB) encoder, a Sparse Heterogeneous Binarization (SHB) decoder that masks homogeneous regions, and a Localized Binarization-aware Mimicking (LBM) distillation loss. Experiments on VideoMatte240K, Distinctions-646, and Adobe Image Matting report state-of-the-art accuracy among 1-bit methods (e.g., 11.82 MAD on VM vs. 28.49 for ReActNet-binarized RVM and 12.82 for BiMatting), with 14.3x FLOPs and 21.6x storage reductions relative to full-precision RVM, plus ARM CPU latency measurements. The paper also presents an information-bottleneck analysis motivating the encoder and decoder designs.

Significance. If the results hold, BiVM is a meaningful step toward practical binarized video matting: it demonstrates that a carefully designed combination of binary-friendly architecture, spatial sparsification, and distillation can recover most of the full-precision accuracy while reducing computation and storage dramatically. The empirical comparison is broad (three datasets, low and high resolution, multiple binarization baselines trained under a shared pipeline) and includes real hardware latency, which strengthens the deployment claim. The paper also ships an ablation study and visual examples. However, the theoretical justification (Theorem 1) has a serious proof inconsistency, and the SHB mask mechanism is underspecified; these points need to be resolved before the claims can be fully credited.

major comments (4)
  1. [Appendix B.1, Theorem 1] The proof of Theorem 1 proves a different statement than the theorem. The theorem defines f(x)=ax+b and considers T=f(X)=aX+b and \hat T=f(sign(X))=a sign(X)+b, but the proof analyzes \hat T=(aX+b) sign(X), which is not the quantity in the theorem. Additionally, the mutual-information computation for continuous X is not rigorous: I(X;T) for an injective deterministic map is infinite, and the bound I(X;\hat T)≤2 log 2 is derived through informal discrete-entropy approximations that are not valid for continuous random variables. Since Theorem 1 is used to motivate the whole information-bottleneck narrative, this needs a correct proof or the theoretical claim must be softened.
  2. [Section 3.4.2, Eq. (16)] The paper never specifies how τ* in Eq. (16) is obtained. It is not stated whether τ* is found by a per-input search over τ, by a closed-form quantile/statistic of the feature map, or by a learned scalar. This is critical for both reproducibility and efficiency: a per-input optimization loop over τ could add non-negligible overhead that is not counted in the reported FLOPs. The entropy H(sign(M(f^{N/16}; m^τ_inc))) is also not fully defined (over which spatial/channel set is the Bernoulli distribution estimated?). Please provide the exact algorithm, its computational cost, and a sensitivity analysis over τ.
  3. [Table 1, Section 4.1.1] The ablation shows that applying SHB to the vanilla baseline collapses MAD to 191.89, and the final 11.82 MAD is reached only when SHB is combined with EBB and LBM. The paper therefore does not support the claim that the entropy-maximized mask by itself identifies matting-relevant regions; it is only useful inside a specific architecture/training combination. The authors should analyze when and why the mask aligns with matting boundaries, and compare SHB against a simpler fixed criterion (e.g., thresholding the difference map without entropy maximization) to isolate the contribution of the entropy objective.
  4. [Table 1 and Section 4.3] The efficiency claim is weakened by the measured latency. In Table 1, the vanilla 1-bit baseline runs at 60.65 ms while BiVM runs at 63.81 ms, despite SHB reducing FLOPs from 0.55G to 0.32G. This suggests the mask computation and sparse execution overhead are non-trivial on ARM CPU. The paper should report latency for the ablations with and without SHB under identical conditions and explicitly discuss the gap between FLOPs and wall-clock time, otherwise the 'efficiency' contribution remains at the level of theoretical FLOPs only.
minor comments (7)
  1. [Figure 1] Figure 1 appears to report BiVM with 12.70 MAD and 0.57 MB, while Table 2 reports 11.82 MAD and 0.67 MB; please reconcile these numbers.
  2. [Eq. (14)] The norm used to compute the difference map in Eq. (14) is not defined; please state whether it is a per-channel L2 norm, an absolute sum, or another operation, since this determines the spatial mask.
  3. [Appendix B.2] The proof of Theorem 2 introduces a variable Y that is not part of the theorem statement; the Markov chain X → Ti → Tj follows directly from Tj being a function of Ti, so the extra variable is confusing and should be removed.
  4. [Section 3.5.1] The sentence 'we apply the complete pre-training phase to all compared binarized video matting networks' is ambiguous: it should clarify whether the baselines are re-trained from scratch on ImageNet with binarization or initialized from the released full-precision weights.
  5. [Abstract and Section 1] The claim 'BiVM surpasses 16.67 on the MAD' should explicitly name the comparison baseline (RVM-ReActNet) in the abstract, as is done later in the introduction, to avoid ambiguity.
  6. [Section 2.3] The citation list contains a stray question mark after reference [91]; please correct.
  7. [Section 3.4.2] The phrase 'decreases the computational FLOPs ... by significant 52.6%' contains a grammar error ('significant' should be 'significantly').

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: BiVM's accuracy/efficiency claims are verified against external benchmarks, and the information-theoretic discussion is motivational rather than load-bearing.

full rationale

The paper's central claims are empirical: BiVM is compared with 1-bit baselines (BNN, DoReFa, ReActNet, ReCU) and with BiMatting on the public VM, D646, and AIM benchmarks under a common training pipeline. No parameter is fitted to the test set, and the SHB threshold tau* is an input-dependent, entropy-maximizing statistic computed from network features (Eq. 16), not a test-set fit. Self-citation of the authors' prior BiMatting paper [70] appears as a compared baseline and as the conference predecessor ('the previous solution proposed in the original conference paper can be seen as a specified baseline method'); the claimed improvement over it is a measured result, not a consequence of the citation. Theorems 1 and 2 restate standard information-theoretic facts (binarization restricts mutual information; composing functions reduces mutual information) and are used only as motivation for shortcuts and the EBB design; they do not supply the numerical accuracy figures. The paper also honestly reports that SHB alone degrades accuracy sharply (MAD 191.89, Table 1), which is a fragility and reproducibility concern rather than circularity. The Section 4.4 limitation explicitly concedes a residual gap to full precision, again without circular dependence. On the evidence in the text, no derivation step reduces to its own inputs by construction, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a standard training pipeline, a heuristic sparsity mask, and a binarization-friendly encoder design. No new physical entities are introduced. The main hand-set choices are the loss weights and shortcut scaling initializations. The most fragile premises are the entropy-maximized mask heuristic and the assertion that grouped convolutions are the main cause of binarized encoder collapse.

free parameters (3)
  • lambda_EBB = 1e-4
    Loss weight for the shortcut regularization in Eq. (13)/(18). Chosen by hand; not fitted to data.
  • lambda_LBM = 1e-4
    Loss weight for the LBM distillation loss in Eq. (21). Chosen by hand.
  • Shortcut scaling initialization = per-layer = 1; cross-layer and block-wise = r x 1e-3, r in [0,1]
    Hand-set initialization for the learnable shortcut scaling factors. Affects optimization dynamics.
assumptions (5)
  • domain assumption Information Bottleneck Markov chain Y to X to T for the matting network.
    Used in Section 3.2.2 to frame the analysis; not proven for video matting networks.
  • standard math Data Processing Inequality (Theorem 2).
    Standard information-theoretic result; proof in Appendix B.2 is essentially correct though the surrounding text about Y is garbled.
  • ad hoc to paper Binarizing grouped/pointwise convolutions causes mutual interference.
    Asserted in Section 3.3.1 without formal proof; motivates replacing grouped convolutions with regular convolutions in EBB.
  • ad hoc to paper Entropy maximization of the binarized masked feature selects a good sparsity threshold.
    Heuristic in Eq. (16); the ablation (Table 1) shows SHB alone collapses accuracy, so the criterion is not independently sufficient.
  • ad hoc to paper Normal distribution assumption for X in Theorem 1.
    Used in the proof; images are not normally distributed, and the proof also changes the definition of T_hat.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BiVM: Accurate Binarized Neural Network for Efficient Video Matting." pith.science (2026). https://pith.science/paper/CMRDNHMW

@misc{pith2026250704456,
  author       = {Pith},
  title        = {Pith review of: BiVM: Accurate Binarized Neural Network for Efficient Video Matting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CMRDNHMW}},
  note         = {Machine review of arXiv:2507.04456}
}
read the original abstract

Deep neural networks for real-time video matting suffer significant computational limitations on edge devices, hindering their adoption in widespread applications such as online conferences and short-form video production. Binarization emerges as one of the most common compression approaches with compact 1-bit parameters and efficient bitwise operations. However, accuracy and efficiency limitations exist in the binarized video matting network due to its degenerated encoder and redundant decoder. Following a theoretical analysis based on the information bottleneck principle, the limitations are mainly caused by the degradation of prediction-relevant information in the intermediate features and the redundant computation in prediction-irrelevant areas. We present BiVM, an accurate and resource-efficient Binarized neural network for Video Matting. First, we present a series of binarized computation structures with elastic shortcuts and evolvable topologies, enabling the constructed encoder backbone to extract high-quality representation from input videos for accurate prediction. Second, we sparse the intermediate feature of the binarized decoder by masking homogeneous parts, allowing the decoder to focus on representation with diverse details while alleviating the computation burden for efficient inference. Furthermore, we construct a localized binarization-aware mimicking framework with the information-guided strategy, prompting matting-related representation in full-precision counterparts to be accurately and fully utilized. Comprehensive experiments show that the proposed BiVM surpasses alternative binarized video matting networks, including state-of-the-art (SOTA) binarization methods, by a substantial margin. Moreover, our BiVM achieves significant savings of 14.3x and 21.6x in computation and storage costs, respectively. We also evaluate BiVM on ARM CPU hardware.

Figures

Figures reproduced from arXiv: 2507.04456 by the authors.

Figure 1
Figure 1. BiVM achieves impressive resource savings, surpassing [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Limitations of the binarized video matting baseline in accuracy and efficiency on the VM dataset. We present (a) the MAD [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Information planes of RVM, RVM-BNN, BiMatting, and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: The overview of BiVM. We apply the Evolvable Binarized Block (EBB) for the encoder, the Sparse Heterogeneous Binarization [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Frame matting comparison among 1-bit RVM-ReActNet, 1-bit BiMatting, 1-bit BiVM (ours), and 32-bit RVM models. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Femporal coherence comparison among 1-bit RVM-ReActNet, 1-bit BiMatting, 1-bit BiVM (ours), and 32-bit RVM models. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 8
Figure 8. Figure 8: Example testing samples. is derived from the initial non-sparse binarized block, which operates at the smallest feature scale, thus generating the mask with minimal computational overhead. This design choice significantly boosts the decoder’s efficiency. In addition, t…
Figure 9
Figure 9. Figure 9: Composite training samples. The last column is the pixels’ temporal standard deviation. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: More visual results. Compared to 1-bit video matting networks using existing binarization methods, our BiVM surpasses [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: Frames in the 9th second of the supplemental video. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

106 extracted references · 63 canonical work pages

  1. [1]

    Universally quan- tized neural compression

    Eirikur Agustsson and Lucas Theis. Universally quan- tized neural compression. Advances in neural information processing systems, 33:12367–12376, 2020

    2020

  2. [2]

    De- signing effective inter-pixel information flow for natural image matting

    Yagiz Aksoy, Tunc Ozan Aydin, and Marc Pollefeys. De- signing effective inter-pixel information flow for natural image matting. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 29–37, 2017

    2017

  3. [3]

    Binarybert: Pushing the limit of bert quantization

    Haoli Bai, Wei Zhang, Lu Hou, Lifeng Shang, Jin Jin, Xin Jiang, Qun Liu, Michael Lyu, and Irwin King. Binarybert: Pushing the limit of bert quantization. In Annual Meeting of the Association for Computational Linguistics , pages 4334– 4348, 2021

    2021

  4. [4]

    A geodesic framework for fast interactive image and video segmentation and matting

    Xue Bai and Guillermo Sapiro. A geodesic framework for fast interactive image and video segmentation and matting. In 2007 IEEE 11th International Conference on Computer Vision, pages 1–8. IEEE, 2007

    2007

  5. [5]

    Proxquant: Quantized neural networks via proximal operators

    Yu Bai, Yu-Xiang Wang, and Edo Liberty. Proxquant: Quantized neural networks via proximal operators. In International Conference on Learning Representations, 2018

    2018

  6. [6]

    Aciq: Analytical clipping for integer quantization of neural networks

    Ron Banner, Yury Nahshan, Elad Hoffer, and Daniel Soudry. Aciq: Analytical clipping for integer quantization of neural networks. 2018

    2018

  7. [7]

    Estimating or propagating gradients through stochastic neurons for conditional computation

    Yoshua Bengio, Nicholas Léonard, and Aaron Courville. Estimating or propagating gradients through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432, 2013

    arXiv 2013

  8. [8]

    Meliusnet: Can binary neural networks achieve mobilenet-level accuracy? arXiv preprint arXiv:2001.05936, 2020

    Joseph Bethge, Christian Bartz, Haojin Yang, Ying Chen, and Christoph Meinel. Meliusnet: Can binary neural networks achieve mobilenet-level accuracy? arXiv preprint arXiv:2001.05936, 2020

    arXiv 2001

Show all 106 references
  1. [9]

    Hierarchical binary cnns for landmark localization with limited re- sources

    Adrian Bulat and Georgios Tzimiropoulos. Hierarchical binary cnns for landmark localization with limited re- sources. IEEE transactions on pattern analysis and machine intelligence, 42(2):343–356, 2018

    2018

  2. [10]

    Xnor-net++: Improved binary neural networks

    Adrian Bulat and Georgios Tzimiropoulos. Xnor-net++: Improved binary neural networks. CoRR, abs/1909.13863, 2019

    1909 arXiv

  3. [11]

    Binarized spectral compres- sive imaging

    Yuanhao Cai, Yuxin Zheng, Jing Lin, Xin Yuan, Yulun Zhang, and Haoqian Wang. Binarized spectral compres- sive imaging. Advances in Neural Information Processing Systems, 36, 2024

    2024

  4. [12]

    Deep learning with low precision by half-wave gaussian quantization

    Zhaowei Cai, Xiaodong He, Jian Sun, and Nuno Vascon- celos. Deep learning with low precision by half-wave gaussian quantization. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 5918–5926, 2017

    2017

  5. [13]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs

    Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L Yuille. Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence , 40(4):834–848, 2017

    2017

  6. [14]

    Rethinking atrous convolu- tion for semantic image segmentation

    Liang-Chieh Chen, George Papandreou, Florian Schroff, and Hartwig Adam. Rethinking atrous convolu- tion for semantic image segmentation. arXiv preprint arXiv:1706.05587, 2017

    2017 arXiv

  7. [15]

    Semantic human matting

    Quan Chen, Tiezheng Ge, Yanyu Xu, Zhiqiang Zhang, Xinxin Yang, and Kun Gai. Semantic human matting. In Proceedings of the 26th ACM international conference on Multimedia, pages 618–626, 2018

    2018

  8. [16]

    Knn matting

    Qifeng Chen, Dingzeyu Li, and Chi-Keung Tang. Knn matting. IEEE transactions on pattern analysis and machine intelligence, 35(9):2175–2188, 2013

    2013

  9. [17]

    Low-bit quantization of neural networks for effi- cient inference

    Yoni Choukroun, Eli Kravchik, Fan Yang, and Pavel Kisilev. Low-bit quantization of neural networks for effi- cient inference. In 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), pages 3009–3018. IEEE, 2019

    2019

  10. [18]

    A bayesian approach to digital matting

    Yung-Yu Chuang, Brian Curless, David H Salesin, and Richard Szeliski. A bayesian approach to digital matting. In Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001 , volume 2, pages II–II. IEEE, 2001

    2001

  11. [19]

    Binaryconnect: Training deep neural networks with binary weights during propagations

    Matthieu Courbariaux, Yoshua Bengio, and Jean-Pierre David. Binaryconnect: Training deep neural networks with binary weights during propagations. Advances in neural information processing systems, 28, 2015

    2015

  12. [20]

    A tutorial on the cross-entropy method

    Pieter-Tjerk De Boer, Dirk P Kroese, Shie Mannor, and Reuven Y Rubinstein. A tutorial on the cross-entropy method. Annals of operations research, 134:19–67, 2005

    2005

  13. [21]

    Regularizing activation distribution for training binarized deep networks

    Ruizhou Ding, Ting-Wu Chin, Zeye Liu, and Diana Mar- culescu. Regularizing activation distribution for training binarized deep networks. In CVPR, June 2019

    2019

  14. [22]

    Hawq: Hessian aware quantization of neural networks with mixed-precision

    Zhen Dong, Zhewei Yao, Amir Gholami, Michael W Mahoney, and Kurt Keutzer. Hawq: Hessian aware quantization of neural networks with mixed-precision. In Proceedings of the IEEE/CVF international conference on computer vision, pages 293–302, 2019

    2019

  15. [23]

    Bitdistiller: Unleashing the potential of sub-4-bit llms via self-distillation

    Dayou Du, Yijia Zhang, Shijie Cao, Jiaqi Guo, Ting Cao, Xiaowen Chu, and Ningyi Xu. Bitdistiller: Unleashing the potential of sub-4-bit llms via self-distillation. arXiv preprint arXiv:2402.10631, 2024

    2024 arXiv

  16. [24]

    Training with quantization noise for extreme model com- pression

    Angela Fan, Pierre Stock, Benjamin Graham, Edouard Grave, Rémi Gribonval, Herve Jegou, and Armand Joulin. Training with quantization noise for extreme model com- pression. arXiv preprint arXiv:2004.07320, 2020

    2004 arXiv

  17. [25]

    Scribble tracker: a matting-based approach for robust tracking

    Jialue Fan, Xiaohui Shen, and Ying Wu. Scribble tracker: a matting-based approach for robust tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence , 34(8):1633–1644, 2011

    2011

  18. [26]

    A cluster sampling method for image matting via sparse coding

    Xiaoxue Feng, Xiaohui Liang, and Zili Zhang. A cluster sampling method for image matting via sparse coding. In Computer Vision–ECCV 2016: 14th European Conference, Am- sterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14, pages 204–219. Springer, 2016

    2016

  19. [27]

    f, b, alpha matting

    Marco Forte and François Pitié. f, b, alpha matting. arXiv preprint arXiv:2003.07711, 2020

    2003 arXiv

  20. [28]

    Shared sampling for real-time alpha matting

    Eduardo SL Gastal and Manuel M Oliveira. Shared sampling for real-time alpha matting. In Computer Graphics Forum, volume 29, pages 575–584. Wiley Online Library, 2010

    2010

  21. [29]

    Larq: An open-source library for training binarized neural networks

    Lukas Geiger and Plumerai Team. Larq: An open-source library for training binarized neural networks. Journal of Open Source Software, 5(45):1746, 2020

    2020

  22. [30]

    A survey of quan- tization methods for efficient neural network inference

    Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. A survey of quan- tization methods for efficient neural network inference. In Low-Power Computer Vision, pages 291–326. Chapman and Hall/CRC, 2022

    2022

  23. [31]

    Random walks for interactive alpha- matting

    Leo Grady, Thomas Schiwietz, Shmuel Aharon, and Rüdiger Westermann. Random walks for interactive alpha- matting. In Proceedings of VIIP, volume 2005, pages 423–

    2005

  24. [32]

    Factormatte: Redefining video matting for re-composition tasks

    Zeqi Gu, Wenqi Xian, Noah Snavely, and Abe Davis. Factormatte: Redefining video matting for re-composition tasks. arXiv preprint arXiv:2211.02145, 2022

    2022 arXiv

  25. [33]

    A global sampling method for alpha matting

    Kaiming He, Christoph Rhemann, Carsten Rother, Xiaoou Tang, and Jian Sun. A global sampling method for alpha matting. In CVPR 2011, pages 2049–2056. Ieee, 2011

    2011

  26. [34]

    Vita: Video instance segmentation via object token association

    Miran Heo, Sukjun Hwang, Seoung Wug Oh, Joon-Young Lee, and Seon Joo Kim. Vita: Video instance segmentation via object token association. Advances in Neural Information Processing Systems, 35:23109–23120, 2022

    2022

  27. [35]

    Context-aware image matting for simultaneous foreground and alpha estimation

    Qiqi Hou and Feng Liu. Context-aware image matting for simultaneous foreground and alpha estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4130–4139, 2019

    2019

  28. [36]

    Searching for mobilenetv3

    Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1314–1324, 2019. JOURNAL ...

    2019

  29. [37]

    Billm: Pushing the limit of post-training quantization for llms

    Wei Huang, Yangdong Liu, Haotong Qin, Ying Li, Shiming Zhang, Xianglong Liu, Michele Magno, and Xiaojuan Qi. Billm: Pushing the limit of post-training quantization for llms. arXiv preprint arXiv:2402.04291, 2024

    2024 arXiv

  30. [38]

    An empirical study of llama3 quantization: From llms to mllms, 2024

    Wei Huang, Xingyu Zheng, Xudong Ma, Haotong Qin, Chengtao Lv, Hong Chen, Jie Luo, Xiaojuan Qi, Xianglong Liu, and Michele Magno. An empirical study of llama3 quantization: From llms to mllms, 2024. URL https://arxiv. org/abs/2404.14047

    2024 arXiv

  31. [39]

    Binarized neural networks

    Itay Hubara, Matthieu Courbariaux, Daniel Soudry, Ran El-Yaniv, and Yoshua Bengio. Binarized neural networks. Advances in neural information processing systems , 29, 2016

    2016

  32. [40]

    Accurate post training quantization with small calibration sets

    Itay Hubara, Yury Nahshan, Yair Hanani, Ron Banner, and Daniel Soudry. Accurate post training quantization with small calibration sets. In International Conference on Machine Learning, pages 4466–4475. PMLR, 2021

    2021

  33. [41]

    A programming language

    Kenneth E Iverson. A programming language. In Proceedings of the May 1-3, 1962, spring joint computer conference, pages 345–351, 1962

    1962

  34. [42]

    Image matting with kl-divergence based sparse sampling

    Levent Karacan, Aykut Erdem, and Erkut Erdem. Image matting with kl-divergence based sparse sampling. In Proceedings of the IEEE international conference on computer vision, pages 424–432, 2015

    2015

  35. [43]

    Mask transfiner for high-quality instance segmentation

    Lei Ke, Martin Danelljan, Xia Li, Yu-Wing Tai, Chi-Keung Tang, and Fisher Yu. Mask transfiner for high-quality instance segmentation. In CVPR, 2022

    2022

  36. [44]

    Video mask transfiner for high-quality video instance segmentation

    Lei Ke, Henghui Ding, Martin Danelljan, Yu-Wing Tai, Chi-Keung Tang, and Fisher Yu. Video mask transfiner for high-quality video instance segmentation. In ECCV, 2022

    2022

  37. [45]

    Is a green screen really necessary for real-time portrait matting? arXiv preprint arXiv:2011.11961, 2020

    Zhanghan Ke, Kaican Li, Yurou Zhou, Qiuhua Wu, Xi- angyu Mao, Qiong Yan, and Rynson WH Lau. Is a green screen really necessary for real-time portrait matting? arXiv preprint arXiv:2011.11961, 2020

    2011 arXiv

  38. [46]

    Modnet: Real-time trimap-free portrait matting via objective decomposition

    Zhanghan Ke, Jiayu Sun, Kaican Li, Qiong Yan, and Rynson WH Lau. Modnet: Real-time trimap-free portrait matting via objective decomposition. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 1140–1147, 2022

    2022

  39. [47]

    A closed- form solution to natural image matting

    Anat Levin, Dani Lischinski, and Yair Weiss. A closed- form solution to natural image matting. IEEE transactions on pattern analysis and machine intelligence , 30(2):228–242, 2007

    2007

  40. [48]

    Vm- former: End-to-end video matting with transformer

    Jiachen Li, Vidit Goel, Marianna Ohanyan, Shant Navasardyan, Yunchao Wei, and Humphrey Shi. Vm- former: End-to-end video matting with transformer. arXiv preprint arXiv:2208.12801, 2022

    2022 arXiv

  41. [49]

    Video- matt: A simple baseline for accessible real-time video matting

    Jiachen Li, Marianna Ohanyan, Vidit Goel, Shant Navasardyan, Yunchao Wei, and Humphrey Shi. Video- matt: A simple baseline for accessible real-time video matting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2177–2186, 2023

    2023

  42. [50]

    Siman: Sign- to-magnitude network binarization

    Mingbao Lin, Rongrong Ji, Zihan Xu, Baochang Zhang, Fei Chao, Chia-Wen Lin, and Ling Shao. Siman: Sign- to-magnitude network binarization. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(5):6277–6288, 2022

    2022

  43. [51]

    Real-time high-resolution background mat- ting

    Shanchuan Lin, Andrey Ryabtsev, Soumyadip Sengupta, Brian L Curless, Steven M Seitz, and Ira Kemelmacher- Shlizerman. Real-time high-resolution background mat- ting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8762–8771, 2021

    2021

  44. [52]

    Backgroundmattingv2, 05 2023

    Shanchuan Lin, Andrey Ryabtsev, Soumyadip Sengupta, Brian L Curless, Steven M Seitz, and Ira Kemelmacher- Shlizerman. Backgroundmattingv2, 05 2023

    2023

  45. [53]

    Robust high-resolution video matting with temporal guidance

    Shanchuan Lin, Linjie Yang, Imran Saleemi, and Soumyadip Sengupta. Robust high-resolution video matting with temporal guidance. In WACV, 2022

    2022

  46. [54]

    Robustvideomatting, 05 2023

    Shanchuan Lin, Linjie Yang, Imran Saleemi, and Soumyadip Sengupta. Robustvideomatting, 05 2023

    2023

  47. [55]

    Neural networks with few multipli- cations

    Zhouhan Lin, Matthieu Courbariaux, Roland Memisevic, and Yoshua Bengio. Neural networks with few multipli- cations. arXiv preprint arXiv:1510.03009, 2015

    2015 arXiv

  48. [56]

    Boosting semantic human matting with coarse annotations

    Jinlin Liu, Yuan Yao, Wendi Hou, Miaomiao Cui, Xu- ansong Xie, Changshui Zhang, and Xian-sheng Hua. Boosting semantic human matting with coarse annotations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8563–8572, 2020

    2020

  49. [57]

    Bit: Robustly binarized multi-distilled trans- former

    Zechun Liu, Barlas Oguz, Aasish Pappu, Lin Xiao, Scott Yih, Meng Li, Raghuraman Krishnamoorthi, and Yashar Mehdad. Bit: Robustly binarized multi-distilled trans- former. Advances in neural information processing systems , 35:14303–14316, 2022

    2022

  50. [58]

    Reactnet: Towards precise binary neural network with generalized activation functions

    Zechun Liu, Zhiqiang Shen, Marios Savvides, and Kwang- Ting Cheng. Reactnet: Towards precise binary neural network with generalized activation functions. In ECCV, 2020

    2020

  51. [59]

    Bi-real net: Enhancing the performance of 1-bit cnns with improved representational capability and advanced training algorithm

    Zechun Liu, Baoyuan Wu, Wenhan Luo, Xin Yang, Wei Liu, and Kwang-Ting Cheng. Bi-real net: Enhancing the performance of 1-bit cnns with improved representational capability and advanced training algorithm. In ECCV, 2018

    2018

  52. [60]

    Training binary neural networks with real-to-binary convolutions

    Brais Martinez, Jing Yang, Adrian Bulat, and Georgios Tzimiropoulos. Training binary neural networks with real-to-binary convolutions. In International Conference on Learning Representations, 2020

    2020

  53. [61]

    Training binary neural networks with real-to-binary convolutions

    Brais Martinez, Jing Yang, Adrian Bulat, and Georgios Tzimiropoulos. Training binary neural networks with real-to-binary convolutions. In ICLR, 2020

    2020

  54. [62]

    Up or down? adaptive rounding for post-training quantization

    Markus Nagel, Rana Ali Amjad, Mart Van Baalen, Christos Louizos, and Tijmen Blankevoort. Up or down? adaptive rounding for post-training quantization. In International Conference on Machine Learning, pages 7197–7206. PMLR, 2020

    2020

  55. [63]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems , 32, 2019

    2019

  56. [64]

    Binarizing mobilenet via evolution-based searching

    Hai Phan, Zechun Liu, Dang Huynh, Marios Savvides, Kwang-Ting Cheng, and Zhiqiang Shen. Binarizing mobilenet via evolution-based searching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13420–13429, 2020

    2020

  57. [65]

    Attention- guided hierarchical structure aggregation for image mat- ting

    Yu Qiao, Yuhao Liu, Xin Yang, Dongsheng Zhou, Min- gliang Xu, Qiang Zhang, and Xiaopeng Wei. Attention- guided hierarchical structure aggregation for image mat- ting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13676–13685, 2020

    2020

  58. [66]

    Bibert: Accurate fully binarized bert

    Haotong Qin, Yifu Ding, Mingyuan Zhang, Qinghua Yan, Aishan Liu, Qingqing Dang, Ziwei Liu, and Xianglong Liu. Bibert: Accurate fully binarized bert. arXiv preprint arXiv:2203.06390, 2022

    2022 arXiv

  59. [67]

    Diverse sample generation: Pushing the limit of generative data-free quantization

    Haotong Qin, Yifu Ding, Xiangguo Zhang, Jiakai Wang, Xianglong Liu, and Jiwen Lu. Diverse sample generation: Pushing the limit of generative data-free quantization. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 45(10):11689–11706, 2023

    2023

  60. [68]

    Binary neural networks: A survey

    Haotong Qin, Ruihao Gong, Xianglong Liu, Xiao Bai, Jingkuan Song, and Nicu Sebe. Binary neural networks: A survey. Pattern Recognition, 2020

    2020

  61. [69]

    Forward and backward information retention for accurate binary neural networks

    Haotong Qin, Ruihao Gong, Xianglong Liu, Mingzhu Shen, Ziran Wei, Fengwei Yu, and Jingkuan Song. Forward and backward information retention for accurate binary neural networks. In CVPR, 2020

    2020

  62. [70]

    Bimatting: Efficient video matting via binarization

    Haotong Qin, Lei Ke, Xudong Ma, Martin Danelljan, Yu- Wing Tai, Chi-Keung Tang, Xianglong Liu, and Fisher Yu. Bimatting: Efficient video matting via binarization. Advances in Neural Information Processing Systems , 36, 2024

    2024

  63. [71]

    Accurate lora-finetuning quanti- zation of llms via information retention

    Haotong Qin, Xudong Ma, Xingyu Zheng, Xiaoyang Li, Yang Zhang, Shouda Liu, Jie Luo, Xianglong Liu, and Michele Magno. Accurate lora-finetuning quanti- zation of llms via information retention. arXiv preprint JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 16 arXiv:24...

    2015 arXiv

  64. [72]

    Bibench: Benchmarking and analyzing network binarization

    Haotong Qin, Mingyuan Zhang, Yifu Ding, Aoyu Li, Ziwei Liu, Fisher Yu, and Xianglong Liu. Bibench: Benchmarking and analyzing network binarization. In ICML, 2023

    2023

  65. [73]

    Distribution-sensitive infor- mation retention for accurate binary neural network

    Haotong Qin, Xiangguo Zhang, Ruihao Gong, Yifu Ding, Yi Xu, and Xianglong Liu. Distribution-sensitive infor- mation retention for accurate binary neural network. International Journal of Computer Vision, 131(1):26–47, 2023

    2023

  66. [74]

    Xnor-net: Imagenet classification using binary convolutional neural networks

    Mohammad Rastegari, Vicente Ordonez, Joseph Redmon, and Ali Farhadi. Xnor-net: Imagenet classification using binary convolutional neural networks. In ECCV, 2016

    2016

  67. [75]

    Principles of neurodynamics

    Frank Rosenblatt. Principles of neurodynamics. percep- trons and the theory of brain mechanisms. Technical report, Cornell Aeronautical Lab Inc Buffalo NY, 1961

    1961

  68. [76]

    Howard, Menglong Zhu, An- drey Zhmoginov, and Liang-Chieh Chen

    Mark Sandler, Andrew G. Howard, Menglong Zhu, An- drey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR, 2018

    2018

  69. [77]

    Back- ground matting: The world is your green screen

    Soumyadip Sengupta, Vivek Jayaram, Brian Curless, Steven M Seitz, and Ira Kemelmacher-Shlizerman. Back- ground matting: The world is your green screen. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2291–2300, 2020

    2020

  70. [78]

    Lipschitz continuity retained binary neural network

    Yuzhang Shang, Dan Xu, Bin Duan, Ziliang Zong, Liqiang Nie, and Yan Yan. Lipschitz continuity retained binary neural network. In ECCV, 2022

    2022

  71. [79]

    Network binarization via contrastive learning

    Yuzhang Shang, Dan Xu, Ziliang Zong, Liqiang Nie, and Yan Yan. Network binarization via contrastive learning. In ECCV, 2022

    2022

  72. [80]

    Information flow in deep neural networks

    Ravid Shwartz-Ziv. Information flow in deep neural networks. arXiv preprint arXiv:2202.06749, 2022

    2022 arXiv

  73. [81]

    Opening the black box of deep neural networks via information

    Ravid Shwartz-Ziv and Naftali Tishby. Opening the black box of deep neural networks via information. arXiv preprint arXiv:1703.00810, 2017

    2017 arXiv

  74. [82]

    The information bottleneck: Theory and applications

    Noam Slonim. The information bottleneck: Theory and applications. PhD thesis, Citeseer, 2002

    2002

  75. [83]

    Deep video matting via spatio-temporal alignment and aggregation

    Yanan Sun, Guanzhi Wang, Qiao Gu, Chi-Keung Tang, and Yu-Wing Tai. Deep video matting via spatio-temporal alignment and aggregation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6975–6984, 2021

    2021

  76. [84]

    Degree-quant: Quantization-aware training for graph neural networks

    Shyam Anil Tailor, Javier Fernandez-Marques, and Nicholas Donald Lane. Degree-quant: Quantization-aware training for graph neural networks. In International Conference on Learning Representations, 2020

    2020

  77. [85]

    Designing bert for convolutional networks: Sparse and hierarchical masked modeling.arXiv preprint arXiv:2301.03580, 2023

    Keyu Tian, Yi Jiang, Qishuai Diao, Chen Lin, Liwei Wang, and Zehuan Yuan. Designing bert for convolutional networks: Sparse and hierarchical masked modeling.arXiv preprint arXiv:2301.03580, 2023

    2023 arXiv

  78. [86]

    Binary graph convolutional network with capacity exploration

    Junfu Wang, Yuanfang Guo, Liang Yang, and Yunhong Wang. Binary graph convolutional network with capacity exploration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  79. [87]

    Gradient matters: Designing binarized neural networks via enhanced information-flow

    Qi Wang, Nianhui Guo, Zhitong Xiong, Zeping Yin, and Xuelong Li. Gradient matters: Designing binarized neural networks via enhanced information-flow. IEEE Transactions on Pattern Analysis and Machine Intelligence , 44(11):7551–7562, 2021

    2021

  80. [88]

    Matting by generation

    Zhixiang Wang, Baiang Li, Jian Wang, Yu-Lun Liu, Jinwei Gu, Yung-Yu Chuang, and Shin’Ichi Satoh. Matting by generation. In ACM SIGGRAPH 2024 Conference Papers , pages 1–11, 2024

    2024

  81. [89]

    Learning channel-wise interactions for binary con- volutional neural networks

    Ziwei Wang, Jiwen Lu, Chenxin Tao, Jie Zhou, and Qi Tian. Learning channel-wise interactions for binary con- volutional neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 568–577, 2019

    2019

  82. [90]

    Learning efficient binarized object detectors with information com- pression

    Ziwei Wang, Jiwen Lu, Ziyi Wu, and Jie Zhou. Learning efficient binarized object detectors with information com- pression. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(6):3082–3095, 2021

    2021

  83. [91]

    Bidet: An efficient binarized object detector

    Ziwei Wang, Ziyi Wu, Jiwen Lu, and Jie Zhou. Bidet: An efficient binarized object detector. In CVPR, 2020

    2020

  84. [92]

    Deep image matting with sparse user interactions

    Tianyi Wei, Dongdong Chen, Wenbo Zhou, Jing Liao, Han- qing Zhao, Weiming Zhang, Gang Hua, and Nenghai Yu. Deep image matting with sparse user interactions. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  85. [93]

    Fast end-to-end trainable guided filter

    Huikai Wu, Shuai Zheng, Junge Zhang, and Kaiqi Huang. Fast end-to-end trainable guided filter. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 1838–1847, 2018

    2018

  86. [94]

    Seqformer: Sequential transformer for video instance segmentation

    Junfeng Wu, Yi Jiang, Song Bai, Wenqing Zhang, and Xiang Bai. Seqformer: Sequential transformer for video instance segmentation. In European Conference on Computer Vision, pages 553–569. Springer, 2022

    2022

  87. [95]

    Deep image matting

    Ning Xu, Brian Price, Scott Cohen, and Thomas Huang. Deep image matting. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 2970–2979, 2017

    2017

  88. [96]

    Poem: 1-bit point-wise operations based on expectation-maximization for efficient point cloud processing

    Sheng Xu, Yanjing Li, Junhe Zhao, Baochang Zhang, and Guodong Guo. Poem: 1-bit point-wise operations based on expectation-maximization for efficient point cloud processing. arXiv preprint arXiv:2111.13386, 2021

    2021 arXiv

  89. [97]

    Recu: Reviving the dead weights in binary neural networks

    Zihan Xu, Mingbao Lin, Jianzhuang Liu, Jie Chen, Ling Shao, Yue Gao, Yonghong Tian, and Rongrong Ji. Recu: Reviving the dead weights in binary neural networks. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5198–5208, 2021

    2021

  90. [98]

    MatAnyone: Stable video matting with consistent memory propagation

    Peiqing Yang, Shangchen Zhou, Jixin Zhao, Qingyi Tao, and Chen Change Loy. MatAnyone: Stable video matting with consistent memory propagation. In CVPR, 2025

    2025

  91. [99]

    Understanding straight- through estimator in training activation quantized neural nets

    Penghang Yin, Jiancheng Lyu, Shuai Zhang, Stanley Osher, Yingyong Qi, and Jack Xin. Understanding straight- through estimator in training activation quantized neural nets. In International Conference on Learning Representations, 2019

    2019

  92. [100]

    Mask guided matting via progressive refinement network

    Qihang Yu, Jianming Zhang, He Zhang, Yilin Wang, Zhe Lin, Ning Xu, Yutong Bai, and Alan Yuille. Mask guided matting via progressive refinement network. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1154–1163, 2021

    2021

  93. [101]

    Diversifying sample generation for accurate data-free quantization

    Xiangguo Zhang, Haotong Qin, Yifu Ding, Ruihao Gong, Qinghua Yan, Renshuai Tao, Yuhang Li, Fengwei Yu, and Xianglong Liu. Diversifying sample generation for accurate data-free quantization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, p...

    2021

  94. [102]

    Attention-guided temporally coherent video object matting

    Yunke Zhang, Chi Wang, Miaomiao Cui, Peiran Ren, Xuan- song Xie, Xian-Sheng Hua, Hujun Bao, Qixing Huang, and Weiwei Xu. Attention-guided temporally coherent video object matting. In Proceedings of the 29th ACM International Conference on Multimedia, pages 5128–5137, 2021

    2021

  95. [103]

    Improving neural network quantization without retraining using outlier channel splitting

    Ritchie Zhao, Yuwei Hu, Jordan Dotzel, Chris De Sa, and Zhiru Zhang. Improving neural network quantization without retraining using outlier channel splitting. In International conference on machine learning , pages 7543–

  96. [104]

    Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients

    Shuchang Zhou, Yuxin Wu, Zekun Ni, Xinyu Zhou, He Wen, and Yuheng Zou. Dorefa-net: Training low bitwidth convolutional neural networks with low bitwidth gradients. CoRR, abs/1606.06160, 2016

    2016 arXiv

  97. [105]

    Towards effective low-bitwidth convolutional neural networks

    Bohan Zhuang, Chunhua Shen, Mingkui Tan, Lingqiao Liu, and Ian D Reid. Towards effective low-bitwidth convolutional neural networks. CVPR, 2018

    2018

  98. [106]

    Extracted Feature

    Dongqing Zou, Xiaowu Chen, Guangying Cao, and Xi- aogang Wang. Unsupervised video matting via sparse and low-rank representation. IEEE transactions on pattern analysis and machine intelligence, 42(6):1501–1514, 2019. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 17...

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.