Pith. sign in

REVIEW 3 major objections 5 minor 71 references

DCVC-MB: Neural B-Frame Video Compression using State Space Models

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A neural video codec that codes B-frames with a Mamba state-space fusion model, beating VTM-19.0 by up to 30% BD-rate.

desk verdict The Mamba B-frame fusion and canonical position embedding are real, well-ablated contributions, but the abstract's headline BD-rate numbers mix base and inference-time variants and one figure matches no variant in Table II, so the SOTA claim needs a reconciled table before it stands. read the letter →

arxiv 2607.14305 v1 pith:AETIX3JL submitted 2026-07-15 cs.CV

Arjun Arora , Calvin-Khang Ta , Carlos Restrepo-Galeano , Kruthi Murali , Naga Akhil E S , Arunkumar Mohananchettiar , Jay Shingala , Tong Shao
show 2 more authors
Peng Yin Sean McCarthy
This is my paper
classification cs.CV
keywords neuralvideocompressionB-framecodingstate-spacemodelsMambabidirectionalfusionentropy-awarelatentskippingcanonicalpositionembedding
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 introduces DCVC-MB, a neural video compression framework that adopts an IBP B-frame structure and uses a bidirectional state-space (Mamba) fusion model to combine information from a past and a future reference frame. The central claim is that this design yields BD-rate savings of up to 30.45% over the VTM-19.0-LDP traditional codec, and is the only neural codec in the comparison to surpass the stronger VTM-19.0-RA configuration (by 1.81%). The authors attribute much of the gain to a canonical position embedding that normalizes coordinates to [0,1], enabling the model to generalize from 256×256 training patches to 720p/1080p/4K, and to an adaptive latent skipping mechanism that reduces entropy coding time by roughly 5–9× without hurting rate-distortion. A sympathetic reader would see this as evidence that B-frame coding, which traditional codecs use for substantial gains, can finally be made competitive in neural codecs at high resolutions.

What carries the argument

The Bidirectional Mamba Block (BMB) is a selective state-space model with two parameter sets — one for the forward row-major scan and one for the backward scan — whose outputs are summed. Each scan discretizes the continuous ODE h'=Ah+Bx with zero-order hold, producing recurrent updates hm = Ā hm−1 + B̄ xm and output ym = C hm. The BMB fuses the feature pyramids of the past and future reference frames to condition the encoder, decoder, and entropy model. The other essential component is the canonical position embedding, which maps token positions to normalized x,y∈[0,1] before applying sinusoidal basis functions, so that the positional harmonics are resolution-independent; this prevents the

What would settle it

Train or fine-tune the model on Vimeo90k at 256×256, then evaluate on HEVC Class B with standard pixel-coordinate positional embeddings; the paper reports a +55.12 BD-rate degradation versus DCVC-DC, while the canonical embedding gives −7.34. A direct falsification test is to evaluate DCVC-MB on sequences with aspect ratios far from training (e.g., vertical 9:16 or ultrawide 21:9) at high resolution; if BD-rate is no better than DCVC-DC, then the canonical embedding does not generalize as claimed. Alternatively, inspect the positional harmonic frequencies at test resolution; if the sampled har

Watch

Extended reading notes

Core claim

DCVC-MB is a neural B-frame codec built on the DCVC-DC backbone but with two decoded picture buffers (past and future) instead of one. The core discovery is that a bidirectional Mamba-based fusion block, running forward and backward row-major scans over concatenated reference features, can effectively exploit temporal redundancy in both directions at linear O(N) complexity. The paper reports BD-rate reductions of up to 8.98% over prior neural codecs (DCVC-FM) and up to 30.45% over VTM-19.0-LDP, and it is the only neural codec in its comparison to exceed VTM-19.0-RA (Inter-GoP=16) by 1.81%. Two inference-time strategies, open GoP and bidirectional coding, provide additive savings, and the ent

Load-bearing premise

The canonical position embedding must preserve the positional harmonics across unseen resolutions and aspect ratios; if the [0,1] normalization does not keep the harmonics aligned at 4K or nonstandard aspect ratios, the reported BD-rate gains over VTM-19.0, which rely on cross-resolution generalization, would shrink substantially.

Editorial extensions

If this is right

  • If the claims hold, neural codecs can now enter the B-frame regime that traditional codecs have exploited for decades, yielding large bitrate savings at comparable quality.
  • The Mamba fusion's linear complexity means B-frame coding fits in 13.8GB VRAM at 1080p, a configuration where transformer-based fusion would exceed 40GB, making high-resolution neural B-frame coding practical on current GPUs.
  • Adaptive latent skipping reduces entropy coding time by ~9× for P-frames and ~5× for B-frames with minor RD cost, which could enable real-time or near-real-time neural coding.
  • The canonical position embedding allows a single checkpoint to be used across resolutions from 256×256 to 4K, eliminating the need for resolution-specific retraining (paper's results show −7.34 BD-rate on HEVC Class B vs DCVC-DC).
  • The two inference-time strategies, open GoP and bidirectional coding, provide additive 0.9%–1% BD-rate gains and can be toggled based on complexity needs; bidirectional coding doubles encoding time but is recommended for storage-critical scenarios.

Reading between the lines

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

  • If the canonical position embedding is truly resolution-agnostic, the same model should handle extreme aspect ratios (e.g., 9:16 vertical or 21:9) without retraining; this is a direct testable consequence not explored in the paper.
  • The adaptive latent skipping, with its per-frame and per-stage relative threshold, is a generic speedup module that could be dropped into other entropy-coded neural codecs (e.g., image compressors) to reduce coding time.
  • The bidirectional Mamba fusion is a generic two-stream context fusion mechanism; it might benefit other tasks requiring past/future context, such as video interpolation or super-resolution, though the paper does not examine these.
  • The paper's YUV420 comparison uses a naive RGB-to-YUV conversion; a model trained directly for YUV420 might close the gap to VTM-RA in that color space, but the paper's claim is limited to RGB BT.709.
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 / 5 minor

Summary. The paper proposes DCVC-MB, a B-frame neural video codec built on the DCVC-DC framework. The main technical contributions are (i) a bidirectional Mamba-based fusion module that combines past and future reference features with linear complexity, (ii) a canonical position embedding designed to generalize across resolutions, (iii) an adaptive latent-skipping mechanism for faster entropy coding, and (iv) two inference-time strategies (open GoP and bidirectional coding). The authors report BD-rate savings over VTM-19.0 and prior neural codecs on HEVC, UVG, and MCL-JCV, and claim state-of-the-art performance. The paper includes ablation studies showing the benefit of Mamba fusion, canonical position embeddings, and the inference-time add-ons, plus runtime and memory analyses.

Significance. If the reported numbers are correct, the work is a relevant empirical contribution to neural B-frame coding: the Mamba-based fusion offers a practical way to use bidirectional references at high resolutions, the canonical position embedding is a simple and potentially generalizable fix for the train/test resolution gap, and the latent-skipping mechanism gives substantial entropy-coding speedups. The ablation study (Table IV) and memory-scaling experiment (Appendix Fig. 1) are valuable. However, the paper does not release code or checkpoints, and the headline quantitative claims are not consistently tied to a single model variant. The numerical inconsistencies described below must be resolved before the SOTA claim can be accepted.

major comments (3)
  1. [Section IV, Table II] The claimed average BD-rate gains are not consistently supported by Table II. The text reports '8.21% and 8.98% over DCVC-FM' for Intra-periods 32 and -1, but at Intra-period=32 no row in Table II gives 8.21%: the base model gives 7.95% (-27.39 vs -19.44), DCVC-MB-og gives 9.18%, and DCVC-MB-bdc-og gives 9.19%. The Introduction's '9.21%' is also unmatched in the table. Please correct these numbers and state precisely which configuration (base, og, bdc, bdc-og) supports each average improvement.
  2. [Section IV.B, Abstract; Section III.G] The headline SOTA claim is misattributed to the base architecture when it actually comes from inference-time variants. The abstract's '1.81% over VTM-19.0-RA' corresponds to DCVC-MB-og (-28.62 vs -26.82); the base model's margin is only 0.57% (-27.39 vs -26.82). Similarly, the '3.43% over DCVC-DC at Intra-period=32' is from DCVC-MB-og (-28.62 vs -25.20), and the '21.86% at Intra-period=-1' is from DCVC-MB-bdc (-30.45 vs -8.59). Since the base model is the architectural contribution and the open-GoP/bidirectional-coding add-ons are separate inference-time choices, every comparison should be labeled with its exact variant.
  3. [Appendix V, Table II; Section IV.A] The canonical position embedding is a load-bearing component for resolution generalization, since training is at 256×256 and testing includes 720p/1080p/4K. The ablation in Appendix Table II covers only HEVC Classes B/C/D (up to 1080p). The main evaluation also includes UVG (4K) and MCL-JCV, but no ablation is reported at those resolutions. To validate the resolution-generalization contribution at the resolutions where the headline results are claimed, please provide an ablation or analysis on UVG (or a 4K subset), or explicitly state that this mechanism's benefit at 4K is not directly measured.
minor comments (5)
  1. [Section III.F, Eq. (2)] The notation in Eq. (2) is unclear: the reader must infer that μ_i and σ_i are the predicted mean and scale of the latent distribution, and that 'mean(σ)' is computed over the stage being coded. Please define these quantities and specify the threshold computation more precisely.
  2. [Appendix II, Table I; Section III.F] The text says the B-frame model is 'approximately 2×' the latency of the P-frame, but the decoder latency in Table I is 492 ms vs 141 ms (≈3.5×), while the encoder is 626 vs 323 (≈1.9×). Please clarify whether the 2× figure refers to encoder only or to a different aggregate.
  3. [Appendix V, Table II] The header 'DCVC-DC p_s nop_s p_ce' is confusing. Please label the columns explicitly (e.g., 'DCVC-DC', 'no pos. emb.', 'standard 2D pos. emb.', 'canonical pos. emb.').
  4. [Abstract, Introduction, Section IV] The percentage improvements differ between the abstract ('up to 8.98%'), the Introduction ('8.98%/9.21%'), and Section IV ('8.21% and 8.98%'). Unify these numbers after correcting the variant attribution.
  5. [General] No code or checkpoints are released. Given the centrality of the SOTA claim and the difficulty of reproducing learned codecs, releasing the trained models and evaluation scripts would substantially increase the credibility of the reported numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: empirical BD-rate comparisons and ablations do not reduce to fitted inputs or self-citations.

full rationale

The paper is an empirical neural codec paper whose central claims are rate-distortion measurements against external baselines (VTM-19.0, DCVC-DC, DCVC-FM). The loss is a standard R-D loss, the fusion/position-embedding/skipping components are evaluated by ablations, and no equation or construction makes a reported prediction equal to a fitted input. The adaptive latent-skipping rule uses the model's own predicted σ, but this is an internal runtime mechanism, not a validation loop for the SOTA claim. The canonical position embedding is a modified standard sinusoidal embedding and its benefit is measured empirically rather than derived by construction. The cited baselines are from other research groups, so there is no load-bearing self-citation chain. Some reported numbers are internally inconsistent or misattributed across model variants (e.g., the 1.81% VTM-RA margin and the 8.21% DCVC-FM cell), but these are reporting/evidence issues, not circularity. Scoring reflects the absence of any step where a 'prediction' reduces to the input by definition.

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

The method introduces no new physical or mathematical entities. It relies on standard SSM discretization, on several domain assumptions about training transfer and metric choice, and on a small number of hand-chosen hyperparameters (λ, Mamba state dimension, number of layers). The latent-skipping threshold is adaptive and not an explicit fitted parameter.

free parameters (4)
  • Lambda (rate-distortion weights) = {85, 170, 380, 840}
    Taken from DCVC-DC/FM; chosen by hand, not fitted in this paper.
  • Mamba state dimension N = Not stated
    Standard Mamba hyperparameter; chosen by hand, affects fusion capacity.
  • Number of BMB layers L = 2
    Chosen by hand; no ablation reported for L.
  • Latent skip threshold = mean(σ) per stage
    Adaptive threshold based on the entropy model's own scale predictions; no explicit fitted value.
assumptions (4)
  • standard math Zero-order hold discretization of the SSM ODE (Eq. 1) yields the discrete update
    Standard result from Mamba; assumed without proof.
  • domain assumption Vimeo90k training distribution transfers to HEVC/UVG/MCL-JCV test sets
    Required for the reported evaluation; the appendix acknowledges domain gaps on animated content.
  • domain assumption The mask M for latent skipping is available at the decoder without side information because σ is predicted by the entropy model
    Assumes encoder/decoder entropy models produce identical σ; not discussed in detail.
  • domain assumption BD-Rate computed with RGB PSNR is an appropriate metric
    Used throughout; YUV420 comparisons are naive per the appendix.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DCVC-MB: Neural B-Frame Video Compression using State Space Models." pith.science (2026). https://pith.science/paper/AETIX3JL

@misc{pith2026260714305,
  author       = {Pith},
  title        = {Pith review of: DCVC-MB: Neural B-Frame Video Compression using State Space Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AETIX3JL}},
  note         = {Machine review of arXiv:2607.14305}
}
abstract

In this paper we propose DCVC-Mamba (DCVC-MB), a neural video codec framework for B-frame coding. Our approach incorporates an IBP frame strategy for low-delay B-frame coding, a spatio-temporal fusion model based on state-space models for bidirectional temporal prediction, and an entropy-aware skipping mechanism that selectively omits coding certain latents to reduce entropy coding times. In addition to our model contributions we also implement two inference-time strategies that enhance compression performance. Experimental evaluation shows that DCVC-MB compares favorably to existing NVCs and traditional codecs. The method demonstrates BD-rate reductions of up to $8.98\%$ on average compared to prior neural video codecs, and improvements of up to $30.45\%$ and $1.81\%$ over the VTM-19.0-LDP and VTM-19.0-RA(Inter-GoP=16) benchmarks, respectively, contributing to advances in neural video compression.

Figures

Figures reproduced from arXiv: 2607.14305 by the authors.

Figure 1
Figure 1. Diagrams showing a) LDP coding using P-frame ordering and b) RA coding using IBP-frame ordering, both in a Closed GoP configuration. Arrows indicate reference relationships. • A novel feature fusion architecture leveraging state-space models that exploits the temporal redundancy of previously decoded frames. • An adaptive entropy based latent-skipping technique for faster entropy coding. • Two inference-time optimiz… view at source ↗
Figure 2
Figure 2. Overview of DCVC-MB architecture, built upon DCVC-DC [ [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Bidirectional Mamba Fusion Model architecture. Left: Frames [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: RD-Curves for Intra-period=32, all frames for HEVC-B, HEVC-C, and HEVC-D. Additional curves in the Appendix. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Subjective quality comparison of DCVC-MB versus DCVC-DC and DCVC-FM. Note the reconstruction of the wrinkles in the player’s jersey which is [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 1
Figure 1. Figure 1: Comparing the inference memory cost of DCVC-MB at varying [PITH_FULL_IMAGE:figures/full_fig_p007_1.png]
Figure 2
Figure 2. Figure 2: Diagram comparing the 0 th sinusoidal harmonic of 2D canonical position embedding vs 2D resolution dependent position embedding at varying resolutions. key difference lies in how the input coordinates are handled: instead of using absolute pixel indices, we normalize s…
Figure 3
Figure 3. Figure 3: Inspecting the frame by frame PSNR vs BPP we see that there are cases where our B-frame model can fail to incorporate the information from the [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 4
Figure 4. Figure 4: RD-Curves for intra-period = 32, All frames [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: RD-Curves for intra-period = -1, All frames [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: RD-Curves for intra-period = 32, 96 frames [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: RD-Curves for intra-period = -1, 96 frames [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: In this example DCVC-MB is able to maintain a higher PSNR and preserve the details where the flower meets the playing card while maintaining an [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 5 linked inside Pith

  1. [1]

    and Ohm, Jens-Rainer , journal=

    Bross, Benjamin and Wang, Ye-Kui and Ye, Yan and Liu, Shan and Chen, Jianle and Sullivan, Gary J. and Ohm, Jens-Rainer , journal=. Overview of the Versatile Video Coding (. 2021 , volume=

    2021

  2. [2]

    and Ohm, Jens-Rainer and Han, Woo-Jin and Wiegand, Thomas , journal=

    Sullivan, Gary J. and Ohm, Jens-Rainer and Han, Woo-Jin and Wiegand, Thomas , journal=. Overview of the High Efficiency Video Coding (. 2012 , volume=

    2012

  3. [3]

    Optics East , year=

    Comparison of the H.263 and H.261 video compression standards , author=. Optics East , year=

  4. [4]

    Communications of the ACM , volume=

    The JPEG still picture compression standard , author=. Communications of the ACM , volume=. 1991 , publisher=

    1991

  5. [5]

    Communications of the ACM , volume=

    MPEG: A video compression standard for multimedia applications , author=. Communications of the ACM , volume=. 1991 , publisher=

    1991

  6. [6]

    ITU SG16 Doc

    Calculation of average PSNR differences between RD-curves , author=. ITU SG16 Doc. VCEG-M33 , year=

  7. [7]

    Joint Video Experts Team , title =

  8. [8]

    2021 , howpublished =

    Joint Video Experts Team , title =. 2021 , howpublished =

    2021

Show all 71 references
  1. [9]

    Sullivan and Ye-Kui Wang , journal=

    Benjamin Bross and Jianle Chen and Jens-Rainer Ohm and Gary J. Sullivan and Ye-Kui Wang , journal=. Developments in International Video Coding Standardization After AVC, With an Overview of Versatile Video Coding (. 2021 , volume=

    2021

  2. [10]

    Christopoulos, Charilaos and Skodras, Athanassios and Ebrahimi, Touradj , journal=. The. 2000 , publisher=

    2000

  3. [11]

    Advanced Video Coding for Generic Audiovisual Services

    H.264 “Advanced Video Coding for Generic Audiovisual Services” , author=

  4. [12]

    arXiv preprint arXiv:2408.08604 , year=

    Bi-Directional Deep Contextual Video Compression , author=. arXiv preprint arXiv:2408.08604 , year=

    arXiv

  5. [13]

    2024 Data Compression Conference (DCC) , year=

    UCVC: A Unified Contextual Video Compression Framework with Joint P-frame and B-frame Coding , author=. 2024 Data Compression Conference (DCC) , year=

    2024

  6. [14]

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

    Learning for video compression with hierarchical quality and recurrent enhancement , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  7. [15]

    2024 IEEE International Conference on Image Processing (ICIP) , year=

    Motion-Adaptive Inference for Flexible Learned B-Frame Compression , author=. 2024 IEEE International Conference on Image Processing (ICIP) , year=

    2024

  8. [16]

    ArXiv , year=

    Hierarchical B-frame Video Coding for Long Group of Pictures , author=. ArXiv , year=

  9. [17]

    1974 , publisher=

    Modern Control Theory , author=. 1974 , publisher=

    1974

  10. [18]

    Long Short-Term Memory , year =

    Sepp Hochreiter and Jürgen Schmidhuber , journal =. Long Short-Term Memory , year =

  11. [19]

    arXiv preprint arXiv:2405.15413 , year=

    MambaVC: Learned Visual Compression with Selective State Spaces , author=. arXiv preprint arXiv:2405.15413 , year=

    arXiv

  12. [20]

    HiPPO: Recurrent Memory with Optimal Polynomial Projections , url =

    Gu, Albert and Dao, Tri and Ermon, Stefano and Rudra, Atri and R\'. HiPPO: Recurrent Memory with Optimal Polynomial Projections , url =. Advances in Neural Information Processing Systems , editor =

  13. [21]

    International Conference on Learning Representations , year=

    Efficiently Modeling Long Sequences with Structured State Spaces , author=. International Conference on Learning Representations , year=

  14. [22]

    European Conference on Computer Vision , pages=

    Long-term temporal context gathering for neural video compression , author=. European Conference on Computer Vision , pages=. 2025 , organization=

    2025

  15. [23]

    and Dao, Tri and Saab, Khaled Kamal and Thomas, Armin W

    Fu, Daniel Y. and Dao, Tri and Saab, Khaled Kamal and Thomas, Armin W. and Rudra, Atri and Ré, Christopher , biburl =. Hungry Hungry Hippos: Towards Language Modeling with State Space Models. , url =. ICLR , crossref =

  16. [24]

    ArXiv , year=

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces , author=. ArXiv , year=

  17. [25]

    Proceedings of the 41st International Conference on Machine Learning , articleno =

    Dao, Tri and Gu, Albert , title =. Proceedings of the 41st International Conference on Machine Learning , articleno =. 2024 , publisher =

    2024

  18. [26]

    European Conference on Computer Vision , year=

    VideoMamba: State Space Model for Efficient Video Understanding , author=. European Conference on Computer Vision , year=

  19. [27]

    ArXiv , year=

    VMamba: Visual State Space Model , author=. ArXiv , year=

  20. [28]

    ArXiv , year=

    Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model , author=. ArXiv , year=

  21. [29]

    arXiv preprint arXiv:2403.15360 , year=

    Simba: Simplified mamba-based architecture for vision and multivariate time series , author=. arXiv preprint arXiv:2403.15360 , year=

    arXiv

  22. [30]

    Advances in Neural Information Processing Systems , volume=

    Deep Contextual Video Compression , author=. Advances in Neural Information Processing Systems , volume=

  23. [31]

    Proceedings of the 30th ACM International Conference on Multimedia , pages=

    Hybrid spatial-temporal entropy modelling for neural video compression , author=. Proceedings of the 30th ACM International Conference on Multimedia , pages=

  24. [32]

    Neural Video Compression with Diverse Contexts , author=

  25. [33]

    Neural Video Compression with Feature Modulation , author=

  26. [34]

    Wang, Guo-Hua and Li, Jiahao and Li, Bin and Lu, Yan , booktitle=

  27. [35]

    2022 , organization=

    Ho, Yung-Han and Chang, Chih-Peng and Chen, Peng-Yu and Gnutti, Alessandro and Peng, Wen-Hsiao , booktitle=. 2022 , organization=

    2022

  28. [36]

    European Conference on Computer Vision , pages=

    Long-term temporal context gathering for neural video compression , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  29. [37]

    Bi-Directional Deep Contextual Video Compression , year=

    Sheng, Xihua and Li, Li and Liu, Dong and Wang, Shiqi , journal=. Bi-Directional Deep Contextual Video Compression , year=

  30. [38]

    , title=

    Frank Bossen et al. , title=. JCTVC-LL1100 , year =

  31. [39]

    Mercat, Alexandre and Viitanen, Marko and Vanne, Jarno , booktitle=

  32. [40]

    2016 , organization=

    Wang, Haiqiang and Gan, Weihao and Hu, Sudeng and Lin, Joe Yuchieh and Jin, Lina and Song, Longguang and Wang, Ping and Katsavounidis, Ioannis and Aaron, Anne and Kuo, C-C Jay , booktitle=. 2016 , organization=

    2016

  33. [41]

    International Journal of Computer Vision (IJCV) , volume=

    Video Enhancement with Task-Oriented Flow , author=. International Journal of Computer Vision (IJCV) , volume=. 2019 , publisher=

    2019

  34. [42]

    2021 , publisher=

    Ma, Di and Zhang, Fan and Bull, David R , journal=. 2021 , publisher=

    2021

  35. [43]

    Lu, Guo and Ouyang, Wanli and Xu, Dong and Zhang, Xiaoyun and Cai, Chunlei and Gao, Zhiyong , booktitle=

  36. [44]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Neural inter-frame compression for video coding , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  37. [45]

    Scale-Space Flow for End-to-End Optimized Video Compression , year=

    Agustsson, Eirikur and Minnen, David and Johnston, Nick and Ballé, Johannes and Hwang, Sung Jin and Toderici, George , booktitle=. Scale-Space Flow for End-to-End Optimized Video Compression , year=

  38. [46]

    European Conference on Computer Vision , pages=

    Alphavc: High-performance and efficient learned video compression , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  39. [47]

    International Conference on Learning Representations , year=

    Transformer-based transform coding , author=. International Conference on Learning Representations , year=

  40. [48]

    2024 Data Compression Conference (DCC) , pages=

    Residual Block Fusion in Low Complexity Neural Network-Based In-loop Filtering for Video Compression , author=. 2024 Data Compression Conference (DCC) , pages=. 2024 , organization=

    2024

  41. [49]

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

    Unified multivariate gaussian mixture for efficient neural image compression , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  42. [50]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Learned image compression with mixed transformer-cnn architectures , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  43. [51]

    IEEE Transactions on Multimedia , volume=

    Temporal context mining for learned video compression , author=. IEEE Transactions on Multimedia , volume=. 2022 , publisher=

    2022

  44. [52]

    International Conference on Learning Representations , year=

    End-to-end Optimized Image Compression , author=. International Conference on Learning Representations , year=

  45. [53]

    arXiv preprint arXiv:1802.01436 , year=

    Variational image compression with a scale hyperprior , author=. arXiv preprint arXiv:1802.01436 , year=

    arXiv

  46. [54]

    2019 IEEE , author=

    Generative adversarial networks for extreme learned image compression. 2019 IEEE , author=. CVF International Conference on Computer Vision (ICCV) , volume=

    2019

  47. [55]

    Advances in neural information processing systems , volume=

    Joint autoregressive and hierarchical priors for learned image compression , author=. Advances in neural information processing systems , volume=

  48. [56]

    Advances in Neural Information Processing Systems , volume=

    Lossy image compression with conditional diffusion models , author=. Advances in Neural Information Processing Systems , volume=

  49. [57]

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

    Image compression with recurrent neural network and generalized divisive normalization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  50. [58]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Towards end-to-end image compression and analysis with transformers , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  51. [59]

    Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

    Full resolution image compression with recurrent neural networks , author=. Proceedings of the IEEE conference on Computer Vision and Pattern Recognition , pages=

  52. [60]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    arXiv

  53. [61]

    IEEE transactions on Computers , volume=

    Discrete cosine transform , author=. IEEE transactions on Computers , volume=. 1974 , publisher=

    1974

  54. [62]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Optical flow estimation using a spatial pyramid network , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  55. [63]

    International Journal of Computer Vision , volume=

    Video enhancement with task-oriented flow , author=. International Journal of Computer Vision , volume=. 2019 , publisher=

    2019

  56. [64]

    On The Computational Complexity of Self-Attention

    Keles, Feyza Duman and Wijewardena, Pruthuvi Mahesakya and Chinmay Hegde. On The Computational Complexity of Self-Attention. Proceedings of Machine Learning Research. 2023

    2023

  57. [65]

    Advances in Neural Information Processing Systems , year=

    Attention is all you need , author=. Advances in Neural Information Processing Systems , year=

  58. [66]

    International Conference on Learning Representations , year=

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , author=. International Conference on Learning Representations , year=

  59. [67]

    ICLR 2023 , url=

    Conditional Positional Encodings for Vision Transformers , author=. ICLR 2023 , url=

    2023

  60. [68]

    Vision Transformer with 2D Explicit Position Encoding , year=

    Li, Yujie and Ma, Zihang and Wang, Xinghe and Wang, Yifu and Tan, Benying , booktitle=. Vision Transformer with 2D Explicit Position Encoding , year=

  61. [69]

    Complexity Analysis Of Next-Generation VVC Encoding And Decoding , year=

    Pakdaman, Farhad and Adelimanesh, Mohammad Ali and Gabbouj, Moncef and Hashemi, Mahmoud Reza , booktitle=. Complexity Analysis Of Next-Generation VVC Encoding And Decoding , year=

  62. [70]

    Advances in Neural Information Processing Systems , volume=

    Nerv: Neural representations for videos , author=. Advances in Neural Information Processing Systems , volume=

  63. [71]

    Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) , month =

    Tang, Chuanbo and Li, Zhuoyuan and Bian, Yifan and Li, Li and Liu, Dong , title =. Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) , month =. 2025 , pages =

    2025

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.