Pith. sign in

REVIEW 3 major objections 6 minor 49 references

ReFrame: Layer Caching for Accelerated Inference in Real-Time Rendering

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Caching reused neural features speeds real-time rendering 1.4x

desk verdict A solid, honest systems paper that adapts DeepCache to rendering networks; the speedup is real but the 'negligible loss' claim needs longer-sequence temporal evidence. read the letter →

arxiv 2506.13814 v1 pith:LXBD6V5B submitted 2025-06-14 cs.GR cs.LGeess.IV

Lufei Liu , Tor M. Aamodt This is my paper · ORCID
classification cs.GRcs.LGeess.IV
keywords layercachingreal-timerenderingtemporalcoherenceencoder-decodernetworksinferenceaccelerationframeextrapolationsupersamplingimagecomposition
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

ReFrame claims that the temporal coherence of real-time rendering—consecutive frames being nearly identical—means much of the neural network work between frames is redundant. The paper shows that in encoder-decoder networks common in rendering pipelines, you can compute the full network on one frame, cache the intermediate features that feed the final decoder block, and on later frames reuse those cached features instead of recomputing the deep layers. This produces average inference speedups around 1.4x, with individual workloads ranging from about 1.05x to 1.85x depending on the cache-refresh sensitivity, and with small measured image-quality losses. The work matters because neural upscaling, frame extrapolation, and image composition are increasingly the bottleneck in real-time graphics, and ReFrame is a training-free way to turn frame-to-frame similarity into lower latency or higher-quality rendering.

What carries the argument

The load-bearing object is the layer cache: a stored copy of the intermediate feature tensor that ordinarily feeds the network's final block, together with a policy deciding when to refresh it. For a U-Net, caching the output of the block before the last one means subsequent frames compute only the first and last blocks, replacing recomputation of all deeper blocks with the saved tensor. ReFrame's contribution is the refresh policy: instead of refreshing every fixed $N$ frames as in DeepCache, it compares the current input to the cached input using symmetric mean absolute percentage error and refreshes only when that delta exceeds a threshold $ au$, with a motion-vector threshold as an alternative when motion data is already available. The cached feature tensor plus the input-delta comparison carries the argument, because the quality/performance trade-off is set entirely by what is cached and when the cache is invalidated.

What would settle it

Run ReFrame for several minutes of interactive play on a mobile or VR device with fast camera cuts and scene transitions, recording worst-case per-frame latency and per-frame image-difference scores. If the 95th-percentile latency is not improved because cache refreshes dominate, or if users can consistently spot quality dips even when average scores stay low, the paper's central trade-off claim is contradicted.

Watch

Extended reading notes

Core claim

The central claim is that intermediate layer outputs in encoder-decoder rendering networks change slowly enough between frames that they can be cached and reused, allowing the network to skip all but the shallowest and deepest blocks on most frames. On the first frame ReFrame stores the encoder's final output that would normally be concatenated with the first-block skip connection; on later frames it concatenates that cached block with the fresh shallow features and runs only the final decoder block, so the deep layers are not recomputed. The paper demonstrates this on three tasks—frame extrapolation, supersampling, and AR image composition—and reports 1.05x to 1.85x speedups, with FLIP losses mostly below 0.05 under the high-sensitivity refresh policy and up to 0.118 FLIP with a mean squared error of 41.40 on supersampling under low sensitivity. It also argues that a frame-delta cache-refresh policy based on symmetric mean absolute percentage error beats fixed every-N refresh because it avoids the sudden quality spikes that users notice, and shows that the saved time can be redirected to higher-quality rendering inputs.

Load-bearing premise

The load-bearing premise is that small average image-difference scores over short test clips of 10 to 20 frames stand in for what users will accept in sustained real-time use; if temporal artifacts accumulate or deployment devices are more sensitive, the central claim of negligible quality loss weakens.

Editorial extensions

If this is right

  • If ReFrame is correct, encoder-decoder inference in rendering pipelines can run about 1.05x to 1.85x faster without retraining, simply by exploiting frame-to-frame similarity.
  • The time saved can be spent on better rendering inputs: the paper shows 3x supersampling with caching beats 4x supersampling without caching in both latency and image quality.
  • Adaptive frame-delta refresh avoids the quality spikes of fixed every-N refresh, making the technique safer for interactive use.
  • ReFrame applies across a range of networks with skip connections and concatenation, including U-Net, U-Net++, and non-standard encoder-decoders such as the Fourier-based supersampling network used in the evaluation.
  • It is independent of delta-based sparse inference methods like DeltaCNN, and combining them compounds the latency reduction, although ReFrame alone already outperforms DeltaCNN in the frame-extrapolation comparison.

Reading between the lines

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

  • The same cache-and-refresh principle should extend to transformer-based rendering networks, since their intermediate activations also change slowly between frames; the paper explicitly leaves transformers out of scope.
  • The benefit will probably depend on hardware: on memory-bound or bandwidth-limited devices, storing and loading large feature tensors may eat much of the FLOP savings, so the measured desktop-GPU speedup is an upper bound until mobile and VR measurements exist.
  • A natural testable extension is a learned or content-aware refresh threshold that anticipates scene changes using motion and scene complexity rather than a fixed SMAPE threshold.
  • Because the saved time can be reallocated to more ray samples or higher input resolution, ReFrame could be framed as a quality-improvement method rather than only a speedup method; the paper demonstrates one such use on 3x supersampling.
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. ReFrame proposes a training-free layer-caching scheme for encoder-decoder style networks (U-Net and U-Net++) used in real-time rendering. At a full-inference frame, the inputs to the final decoder block are cached; in subsequent frames, those cached features are reused and only the shallowest encoder block and the final block are computed. The paper evaluates several cache refresh policies: the DeepCache every-N schedule, a non-linear schedule, a motion-vector threshold, and a proposed SMAPE-based frame-delta policy with high and low sensitivity settings. Experiments on three workloads (ExtraNet for frame extrapolation, FBSR for neural supersampling, and Implicit Depth for image composition) report speedups between 1.05x and 1.85x, with FLIP/LPIPS/SSIM/PSNR/MSE measured relative to the baseline network. The central claim is 1.4x average inference speedup with negligible quality loss.

Significance. If the claims hold, ReFrame is a practical, retraining-free way to cut encoder-decoder inference latency in real-time rendering pipelines, with direct applicability to the U-Net-style networks that dominate current denoising, supersampling, and frame-generation workloads. The paper is commendably transparent: it reports per-scene results, ablates cache depth and refresh policies, includes a null-hypothesis test on cache contents, gives latency percentiles, releases code, and explicitly discusses limitations. The frame-delta policy (Delta H) shows genuinely small quality degradation in the reported clips. However, the headline result mixes two operating points, and the evidence for "negligible quality loss" over sustained use is thinner than the abstract implies; both issues are addressable in revision.

major comments (3)
  1. [Abstract and Table 1] The headline "1.4x speedup on average with negligible quality loss" is not tied to a single operating point. In Table 1, the high-sensitivity policy (Delta H) averages about 1.22x speedup (FLIP up to 0.049), while the low-sensitivity policy (Delta L) averages about 1.56x speedup but reaches FLIP 0.118 and MSE 41.40 on supersampling. The abstract should either report the speedup at the Delta H operating point, or explicitly state that the 1.4x figure is the average across two sensitivity settings and that "negligible loss" applies only to the high-sensitivity setting. As written, the claim implies that 1.4x speedup is achievable with negligible loss, which the table does not support.
  2. [Section 4.2, Tables 1 and 2] The "negligible quality loss" argument relies on comparing FLIP scores from Table 1 to an acceptable-loss band of 0.05-0.28 taken from other neural rendering systems, but the two sets of numbers are not commensurate. Table 1 reports FLIP relative to the baseline network output, whereas the cited systems (Müller et al., Li et al., Vaidyanathan et al.) report absolute quality against ground truth. For the same scene, Table 2 shows Delta L at FLIP 0.401 vs ground truth compared to the baseline's 0.398, i.e., an added error of 0.003, while Table 1 lists Delta L at FLIP 0.118 relative to baseline. The manuscript must state explicitly which reference is used in each table and justify comparing a relative error metric to an absolute quality bar.
  3. [Appendix A.7, Tables 6 and 10, Figure 9] The evaluation does not adequately demonstrate that cache refreshes do not cause noticeable temporal artifacts in sustained use. The main test sequences are 10-20 frames (Table 6), the policy ablation uses 10 frames (Table 5), and the DeltaCNN comparison uses 3 frames (Table 11). The "longer sequences" in Table 10 report only aggregate metrics, so they cannot reveal per-frame spikes at refresh points. Figure 9, which shows per-frame MAPE over 100 frames for Implicit Depth, is a step in the right direction but covers only one network. To support the central quality claim, please provide per-frame quality traces (e.g., FLIP or error vs frame index) over a sustained sequence for each of the three tasks, or otherwise bound the refresh artifact explicitly.
minor comments (6)
  1. [Section 3.2.2] The notation "SM AP E(I, It)" is garbled; please fix the spacing and define SMAPE formally.
  2. [Section 4.2.2] The sentence "3× upscaling with a cache achieves a higher FLIP score with the Delta H policy" appears to be the opposite of the intended meaning; Table 2 shows the 3x Delta H FLIP is essentially equal to the 3x baseline, and the surrounding text discusses a quality improvement, so "higher" should be "lower" or "comparable."
  3. [Table 5 caption] The caption says "Delta L results in better image quality with fewer cache refreshes," but Delta L has 3 refreshes while N-5 has 2; please clarify the comparison basis (e.g., better quality at a comparable or lower refresh count than other policies for the same quality level).
  4. [Section 4.3] The "small lag during cache refreshes to compute frame deltas" is described qualitatively; please report the measured overhead in milliseconds or point to a table that quantifies it.
  5. [Appendix A.3, Table 7] The SMAPE thresholds τ = 0.20 and τ = 0.25 are listed without units or interpretation; please state that these are percentages (i.e., 20% and 25%) and briefly explain how they relate to the per-pixel deltas described in the text.
  6. [Figure 13] The optical-flow comparison figure has axis labels that may be difficult to read in print; please ensure the legend and axes are legible at the final page size.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the caching mechanism is explicitly attributed to DeepCache, the policy thresholds are disclosed hyperparameters, and all quality and latency claims are measured against external baselines.

full rationale

ReFrame is an empirical systems paper, not a derivation, and its central claim does not reduce to its own inputs by construction. The caching equations (Eqs. 2-4 in Section 3.1) are explicitly attributed to DeepCache (Ma et al., 2024), an external citation, and the speedup and quality results are measured against baseline network outputs rather than implied by the cache definition. The SMAPE thresholds (Table 7) are disclosed hyperparameters; the Delta H and Delta L configurations are evaluated operating points, not fitted parameters later renamed as predictions, and Table 5 directly ablates policy choice. The 'negligible quality loss' interpretation is anchored to an acceptable FLIP band (0.05-0.28) borrowed from external neural-rendering systems (Muller et al. 2021, Li et al. 2022, Vaidyanathan et al. 2023), not to any self-derived criterion. Appendix A.5's null-hypothesis test independently checks that the cached features are content-bearing rather than interchangeable with arbitrary values. The paper also honestly discloses that ReFrame reduces average latency but cannot maintain a consistently faster frame rate (Section 5, Table 12) and that longer sequences do not change the main results (Appendix A.7). The reader-identified concerns about short test clips and mobile-device latency transfer are external validity limitations, not circularity. No load-bearing step is justified solely by self-citation, and no prediction is forced by definition or by a fitted input.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central trade-off claim depends on three hand-set thresholds (two SMAPE settings and one motion threshold), a cache-depth choice, and the domain assumption that temporal coherence persists in deep features. No new physical or architectural entities are proposed; the cache is a memory structure. The paper discloses all of these, and the ablation in Table 5 compares the policies, so the ledger is small and mostly honest.

free parameters (5)
  • SMAPE refresh threshold tau (Delta H) = 0.20
    Hand-set threshold on input image change that triggers a full cache refresh; directly controls the quality and latency trade-off (Section 3.2.2, Appendix A.3 Table 7).
  • SMAPE refresh threshold tau (Delta L) = 0.25
    Lower-sensitivity variant that caches longer and produces the larger speedups and larger quality losses in Table 1.
  • Motion vector threshold tau = 1
    Hand-set average-motion threshold for the motion-vector policy compared in the ablation (Appendix A.3 Table 7).
  • Non-linear policy parameters c and p = c = 110, p = 1.4
    DeepCache's non-linear refresh schedule parameters, reused as a baseline comparison (Appendix A.3 Table 7).
  • Cache depth and U-Net++ configuration = Level 3 / Config B
    The ablation in Table 4 chooses the deepest cache point for maximum FLOPs reduction; cache depth is itself a tunable affecting the quality and performance trade-off.
assumptions (4)
  • domain assumption Consecutive frames in real-time rendering are highly correlated, and this correlation persists in deep encoder features.
    The core premise of the method, argued visually in Figure 1 and used in Section 3.1 to justify reusing cached features across frames.
  • standard math The compositional block model of U-Net and U-Net++ in equations (1) to (4) matches the actual tested networks, so substituting the cached tensor C_t for the full encoder output preserves network semantics.
    The derivation of which blocks can be skipped relies on the concatenation structure described in Section 3.1.
  • domain assumption FLIP and LPIPS measured against the baseline output are valid proxies for user-perceived quality in this setting, and the acceptable-loss ranges cited from other neural rendering systems transfer here.
    Used to label quality loss negligible in Section 4.2; the Delta L supersampling point (FLIP 0.118) depends on this bar.
  • domain assumption A cache produced by the same network on an earlier frame, injected at the final decoder block, does not require retraining or adaptation for the network to produce stable outputs across frames.
    The method is training-free; the null-hypothesis test in Appendix A.5 is the only supporting evidence, showing cache contents matter and small noise is tolerated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ReFrame: Layer Caching for Accelerated Inference in Real-Time Rendering." pith.science (2026). https://pith.science/paper/LXBD6V5B

@misc{pith2026250613814,
  author       = {Pith},
  title        = {Pith review of: ReFrame: Layer Caching for Accelerated Inference in Real-Time Rendering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LXBD6V5B}},
  note         = {Machine review of arXiv:2506.13814}
}
read the original abstract

Graphics rendering applications increasingly leverage neural networks in tasks such as denoising, supersampling, and frame extrapolation to improve image quality while maintaining frame rates. The temporal coherence inherent in these tasks presents an opportunity to reuse intermediate results from previous frames and avoid redundant computations. Recent work has shown that caching intermediate features to be reused in subsequent inferences is an effective method to reduce latency in diffusion models. We extend this idea to real-time rendering and present ReFrame, which explores different caching policies to optimize trade-offs between quality and performance in rendering workloads. ReFrame can be applied to a variety of encoder-decoder style networks commonly found in rendering pipelines. Experimental results show that we achieve 1.4x speedup on average with negligible quality loss in three real-time rendering tasks. Code available: https://ubc-aamodt-group.github.io/reframe-layer-caching/

Figures

Figures reproduced from arXiv: 2506.13814 by the authors.

Figure 1
Figure 1. Input similarity in a rendering workload showing every three frames. Upper right plot shows the similarity compared to Frame 0 (Frame t vs. Frame 0) and similarity between consecutive frames (Frame t vs. Frame t − 1) measured as FLIP scores. approximate features in the cache used by ReFrame must trade image quality for improved frame rates in a way that does not impede the user experience. Furthermore, diffu￾sion mo… view at source ↗
Figure 2
Figure 2. Diagrams of U-Net (a) and U-Net++ (b) architectures with layer caching. In frame t, the network computes the full inference end-to-end, saving intermediate outputs into a cache. In subsequent frames t + 1, the network reuses the cached results instead of re-computing the intermediate layers. In some cases, ReFrame can be applied on networks beyond U-Net and U-Net++ architectures if the network comprises concatenatio… view at source ↗
Figure 3
Figure 3. shows the features at different depths (levels) over consecutive frames in a rendering workload. Caching the last block is the most effective for reducing the number of FLOPs required, but also suffers the highest quality degra￾dation because the changes in intermediate features (light green regions in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Frame extrapolation results with and without caching on the Asian Village scene using ExtraNet. The FLIP error map is shown on the right (pink indicates highly noticeable regions of differences). 4.2. Performance and Quality Results [PITH_FULL_IMAGE:figures/full_fig_p…
Figure 5
Figure 5. Figure 5: Supersampling results on the Sun Temple scene using FBSR. Ground truth image is rendered at high resolution. Delta H and Delta L show results with caching enabled. 3× supersampling shows quality improvements from using a larger input image. Zoomed-in regions show detai…
Figure 6
Figure 6. Figure 6: Implicit Depth results combining Real and Render into Composite. The FLIP error map is shown on the right comparing results with and without using a cache, where pink indicates highly noticeable regions of differences. 0 20 40 60 80 100 Frame 0 10 20 Inference Time (ms…
Figure 7
Figure 7. Figure 7: Inference time of decoder blocks in Implicit Depth over 100 frames. Both N-5 and Delta L policies show significant reduc￾tions in inference time compared to No Cache. Delta L spreads cache refreshes more strategically to enforce output quality. Frame Theoretical FLOPs …
Figure 8
Figure 8. Figure 8: Ablation study results for ExtraNet U-Net architecture. Config A caches all U-Net++ blocks, while Config B caches only the last block. Policy Delta L Delta H N-5 N-2 Motion Vector Non-Linear N-5 # Refresh Frames 3 6 2 5 4 2 FLIP ↓ 0.035 0.021 0.040 0.025 0.030 0.040 SS…
Figure 10
Figure 10. Figure 10: Diagrams of the FBSR architecture with ReFrame. In frame t, the network computes the full inference end-to-end, saving intermediate features from the temporal recurrent feature extractor and HR G-buffer feature extractor. In subsequent frames t + 1, the network reuses…
Figure 11
Figure 11. Figure 11: shows the Level 1-3 cache configurations used in the ablation study for U-Net architectures [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Ablation configurations for Implicit Depth U-Net++ architecture [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Optical flow motion comparison matching our test scenes to real-world game play from GamingVideoSET (Barman et al., 2018) and CGVSD (Zadtootaghaj et al., 2020). Ref 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Frame 0.0 0.1 0.2 0.3 0.4 FLIP Score (Rel. to Ref Frame…
Figure 14
Figure 14. Figure 14: Null hypothesis test with Sun Temple on ExtraNet. video compression algorithms and contain a representative variety of video game content. We measure the per-pixel deltas between each subsequent frame, as we use this metric to trigger cache refreshes. We also measure …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 47 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    o ller, T., Oskarsson, M., str \

    Andersson, P., Nilsson, J., Akenine-M \"o ller, T., Oskarsson, M., str \"o m, K., and Fairchild, M. D. FLIP : A difference evaluator for alternating images. Proc. Int'l Conf. on Computer Graphics and Interactive Techniques (SIGGRAPH), 3 0 (2), 2020

    work page 2020

  3. [3]

    Parallel frame rendering: Trading responsiveness for energy on a mobile gpu

    Arnau, J.-M., Parcerisa, J.-M., and Xekalakis, P. Parallel frame rendering: Trading responsiveness for energy on a mobile gpu. In Proc. IEEE/ACM Conf. on Par. Arch. and Comp. Tech. (PACT), pp.\ 83--92, 2013

    work page 2013

  4. [4]

    G., and M \"o ller, S

    Barman, N., Zadtootaghaj, S., Schmidt, S., Martini, M. G., and M \"o ller, S. GamingVideoSET : a dataset for gaming video streaming applications. In Workshop on Network and Systems Support for Games (NetGames). IEEE, 2018

    work page 2018

  5. [5]

    T., Gowda, M., and Das, C

    Bhuyan, S., Ying, Z., Kandemir, M. T., Gowda, M., and Das, C. R. GameStreamSR : Enabling neural-augmented game streaming on commodity mobile platforms. In Proc. IEEE/ACM Int'l Symp. on Computer Architecture (ISCA), 2024

    work page 2024

  6. [6]

    Interactive neural cascade denoising for 1-spp Monte Carlo images

    Chen, Y., Lu, Y., Zhang, X., and Xie, N. Interactive neural cascade denoising for 1-spp Monte Carlo images. The Visual Computer, 39 0 (8), 2023

    work page 2023

  7. [7]

    Online neural denoising with cross-regression for interactive rendering

    Choi, H., Hong, S., Ha, I., Kang, N., and Moon, B. Online neural denoising with cross-regression for interactive rendering. ACM Transactions on Graphics (TOG), 43 0 (6), 2024

    work page 2024

  8. [8]

    R., de Boer, G

    Chowdhury, H., Kawiak, R. R., de Boer, G. F., and Xavier, L. Intel XeSS -an AI based super sampling solution for real-time rendering. In Game Developers Conference (GDC), volume 1, pp.\ 7, 2022

    work page 2022

Show all 49 references
  1. [9]

    L., Zaremba, W., Bruna, J., LeCun, Y., and Fergus, R

    Denton, E. L., Zaremba, W., Bruna, J., LeCun, Y., and Fergus, R. Exploiting linear structure within convolutional networks for efficient evaluation. Proc. Conf. and Workshop on Neural Information Processing Systems (NeurIPS), 27, 2014

    2014

  2. [10]

    Event neural networks

    Dutson, M., Li, Y., and Gupta, M. Event neural networks. In European Conf. on Computer Vision (ECCV), 2022

    2022

  3. [11]

    Eventful transformers: leveraging temporal redundancy in vision transformers

    Dutson, M., Li, Y., and Gupta, M. Eventful transformers: leveraging temporal redundancy in vision transformers. In Proc. IEEE Int'l Conf. on Computer Vision (ICCV), 2023

    2023

  4. [12]

    ExtraNet : real-time extrapolated rendering for low-latency temporal supersampling

    Guo, J., Fu, X., Lin, L., Ma, H., Guo, Y., Liu, S., and Yan, L.-Q. ExtraNet : real-time extrapolated rendering for low-latency temporal supersampling. ACM Transactions on Graphics (TOG), 40 0 (6), 2021

    2021

  5. [13]

    Han, S., Mao, H., and Dally, W. J. Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. Proc. Int'l Conf. on Learning Representations (ICLR), 2016

    2016

  6. [14]

    Low-latency space-time supersampling for real-time rendering

    He, R., Zhou, S., Sun, Y., Cheng, R., Tan, W., and Yan, B. Low-latency space-time supersampling for real-time rendering. In Proc. Conf. on Artificial Intelligence (AAAI), volume 38, 2024

    2024

  7. [15]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Proc. Conf. and Workshop on Neural Information Processing Systems (NeurIPS), 33, 2020

    2020

  8. [16]

    Binarized neural networks

    Hubara, I., Courbariaux, M., Soudry, D., El-Yaniv, R., and Bengio, Y. Binarized neural networks. Proc. Conf. and Workshop on Neural Information Processing Systems (NeurIPS), 29, 2016

    2016

  9. [17]

    Cambricon-D : Full-network differential acceleration for diffusion models

    Kong, W., Hao, Y., Guo, Q., Zhao, Y., Song, X., Li, X., Zou, M., Du, Z., Zhang, R., Liu, C., et al. Cambricon-D : Full-network differential acceleration for diffusion models. In Proc. IEEE/ACM Int'l Symp. on Computer Architecture (ISCA), 2024

    2024

  10. [18]

    Optimal brain damage

    LeCun, Y., Denker, J., and Solla, S. Optimal brain damage. Advances in neural information processing systems, 2, 1989

    1989

  11. [19]

    Neural 3D video synthesis from multi-view video

    Li, T., Slavcheva, M., Zollhoefer, M., Green, S., Lassner, C., Kim, C., Schmidt, T., Lovegrove, S., Goesele, M., Newcombe, R., et al. Neural 3D video synthesis from multi-view video. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  12. [20]

    and Burnes, A

    Lin, H. and Burnes, A. NVIDIA DLSS 4 Introduces Multi Frame Generation & Enhancements For All DLSS Technologies , 2025. URL https://www.nvidia.com/en-us/geforce/news/dlss4-multi-frame-generation-ai-innovations/

    2025

  13. [21]

    DeepCache : Accelerating diffusion models for free

    Ma, X., Fang, G., and Wang, X. DeepCache : Accelerating diffusion models for free. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  14. [22]

    Diffy: A d \'e j \`a vu-free differential deep neural network accelerator

    Mahmoud, M., Siu, K., and Moshovos, A. Diffy: A d \'e j \`a vu-free differential deep neural network accelerator. In Proc. IEEE/ACM Symp. on Microarch. (MICRO), 2018

    2018

  15. [23]

    Efficient neural supersampling on a novel gaming dataset

    Mercier, A., Erasmus, R., Savani, Y., Dhingra, M., Porikli, F., and Berger, G. Efficient neural supersampling on a novel gaming dataset. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pp.\ 296--306, 2023

    2023

  16. [24]

    Real-time neural radiance caching for path tracing

    M \"u ller, T., Rousselle, F., Nov \'a k, J., and Keller, A. Real-time neural radiance caching for path tracing. ACM Transactions on Graphics (TOG), 40 0 (4), 2021

    2021

  17. [25]

    D., Keskin, C., Wang, R., and Steinberger, M

    Parger, M., Tang, C., Twigg, C. D., Keskin, C., Wang, R., and Steinberger, M. DeltaCNN : End-to-end CNN inference of sparse frame differences in videos. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  18. [26]

    D., Keskin, C., Wang, R., and Steinberger, M

    Parger, M., Tang, C., Neff, T., Twigg, C. D., Keskin, C., Wang, R., and Steinberger, M. MotionDeltaCNN : Sparse CNN inference of frame differences in moving camera videos with spherical buffers and padded convolutions. In Proc. IEEE Int'l Conf. on Computer Vision (ICCV), 2023

    2023

  19. [27]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  20. [28]

    U-net: Convolutional networks for biomedical image segmentation

    Ronneberger, O., Fischer, P., and Brox, T. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention (MICCAI), pp.\ 234--241, 2015

    2015

  21. [29]

    SimpleRecon : 3D reconstruction without 3D convolutions

    Sayed, M., Gibson, J., Watson, J., Prisacariu, V., Firman, M., and Godard, C. SimpleRecon : 3D reconstruction without 3D convolutions. In European Conf. on Computer Vision (ECCV), 2022

    2022

  22. [30]

    Scardigli, A., Cavigelli, L., and M \"u ller, L. K. RL -based stateful neural adaptive sampling and denoising for real-time path tracing. Proc. Conf. and Workshop on Neural Information Processing Systems (NeurIPS), 36, 2024

    2024

  23. [31]

    V., Wimmer, M., and Eisemann, E

    Scherzer, D., Yang, L., Mattausch, O., Nehab, D., Sander, P. V., Wimmer, M., and Eisemann, E. Temporal coherence methods in real-time rendering. In Computer Graphics Forum, volume 31, 2012

    2012

  24. [32]

    Frustum volume caching for accelerated NeRF rendering

    Steiner, M., K \"o hler, T., Radl, L., and Steinberger, M. Frustum volume caching for accelerated NeRF rendering. Proc. Int'l Conf. on Computer Graphics and Interactive Techniques (SIGGRAPH), 7 0 (3), 2024

    2024

  25. [33]

    FlexCache : Flexible approximate cache system for video diffusion

    Sun, D., Tian, H., Lu, T., and Liu, S. FlexCache : Flexible approximate cache system for video diffusion. arXiv preprint arXiv:2501.04012, 2024

    2024 arXiv

  26. [34]

    QoE -based cross-layer optimization of wireless video with unperceivable temporal video quality fluctuation

    Thakolsri, S., Kellerer, W., and Steinbach, E. QoE -based cross-layer optimization of wireless video with unperceivable temporal video quality fluctuation. In IEEE Int'l Conf. on communications (ICC), 2011

    2011

  27. [35]

    Random-access neural compression of material textures

    Vaidyanathan, K., Salvi, M., Wronski, B., Akenine-Moller, T., Ebelin, P., and Lefohn, A. Random-access neural compression of material textures. ACM Transactions on Graphics (TOG), 42 0 (4), 2023

    2023

  28. [36]

    Interactive rendering using the render cache

    Walter, B., Drettakis, G., and Parker, S. Interactive rendering using the render cache. In Rendering Techniques' 99: Proc. of the Eurographics Workshop, pp.\ 19--30. Springer, 1999

    1999

  29. [37]

    J., Vicente, S., Mac Aodha, O., and Firman, M

    Watson, J., Sayed, M., Qureshi, Z., Brostow, G. J., Vicente, S., Mac Aodha, O., and Firman, M. Virtual occlusions through implicit depth. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pp.\ 9053--9064, 2023

    2023

  30. [38]

    Cache me if you can: Accelerating diffusion models through block caching

    Wimbauer, F., Wu, B., Schoenfeld, E., Dai, X., Hou, J., He, Z., Sanakoyeu, A., Zhang, P., Tsai, S., Kohler, J., et al. Cache me if you can: Accelerating diffusion models through block caching. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  31. [39]

    ExtraSS : A framework for joint spatial super sampling and frame extrapolation

    Wu, S., Kim, S., Zeng, Z., Vembar, D., Jha, S., Kaplanyan, A., and Yan, L.-Q. ExtraSS : A framework for joint spatial super sampling and frame extrapolation. In Proc. Int'l Conf. on Computer Graphics and Interactive Techniques in Asia (SIGGRAPH Asia), 2023

    2023

  32. [40]

    Neural supersampling for real-time rendering

    Xiao, L., Nouri, S., Chapman, M., Fix, A., Lanman, D., and Kaplanyan, A. Neural supersampling for real-time rendering. ACM Transactions on Graphics (TOG), 39 0 (4), 2020

    2020

  33. [41]

    X., and Liu, X

    Xu, M., Zhu, M., Liu, Y., Lin, F. X., and Liu, X. Deepcache: Principled cache for mobile deep vision. In Proc. Int'l Conf. on mobile computing and networking (MobiCom), pp.\ 129--144, 2018

    2018

  34. [42]

    MNSS : Neural supersampling framework for real-time rendering on mobile devices

    Yang, S., Zhao, Y., Luo, Y., Wang, H., Sun, H., Li, C., Cai, B., and Jin, X. MNSS : Neural supersampling framework for real-time rendering on mobile devices. IEEE Transactions on Visualization and Computer Graphics (TVCG), 2023

    2023

  35. [43]

    Mob-FGSR : Frame generation and super resolution for mobile real-time rendering

    Yang, S., Zhu, Q., Zhuge, J., Qiu, Q., Li, C., Yan, Y., Xu, H., Yan, L.-Q., and Jin, X. Mob-FGSR : Frame generation and super resolution for mobile real-time rendering. In Proc. Int'l Conf. on Computer Graphics and Interactive Techniques (SIGGRAPH), 2024

    2024

  36. [44]

    ShadowMover : Automatically projecting real shadows onto virtual object

    Yu, P., Guo, J., Huang, F., Chen, Z., Wang, C., Zhang, Y., and Guo, Y. ShadowMover : Automatically projecting real shadows onto virtual object. IEEE Transactions on Visualization and Computer Graphics (TVCG), 29 0 (5), 2023

    2023

  37. [45]

    S., M \"o ller, S., and Griwodz, C

    Zadtootaghaj, S., Schmidt, S., Sabet, S. S., M \"o ller, S., and Griwodz, C. Quality estimation models for gaming video streaming services using perceptual video quality dimensions. In Proceedings of the ACM multimedia systems conference, 2020

    2020

  38. [46]

    Deep Fourier -based arbitrary-scale super-resolution for real-time rendering

    Zhang, H., Guo, J., Zhang, J., Qin, H., Feng, Z., Yang, M., and Guo, Y. Deep Fourier -based arbitrary-scale super-resolution for real-time rendering. In Proc. Int'l Conf. on Computer Graphics and Interactive Techniques (SIGGRAPH), 2024

    2024

  39. [47]

    A., Shechtman, E., and Wang, O

    Zhang, R., Isola, P., Efros, A. A., Shechtman, E., and Wang, O. The unreasonable effectiveness of deep features as a perceptual metric. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pp.\ 586--595, 2018

    2018

  40. [48]

    FuseSR : Super resolution for real-time rendering through efficient multi-resolution fusion

    Zhong, Z., Zhu, J., Dai, Y., Zheng, C., Chen, G., Huo, Y., Bao, H., and Wang, R. FuseSR : Super resolution for real-time rendering through efficient multi-resolution fusion. In Proc. Int'l Conf. on Computer Graphics and Interactive Techniques in Asia (SIGGRAPH Asia), 2023

    2023

  41. [49]

    M., Tajbakhsh, N., and Liang, J

    Zhou, Z., Rahman Siddiquee, M. M., Tajbakhsh, N., and Liang, J. UNet++ : A nested U-net architecture for medical image segmentation. In Medical image computing and computer-assisted intervention (MICCAI). Springer, 2018

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.