Pith. sign in

REVIEW 5 major objections 6 minor 60 references

Amulet: Frame Extrapolation Through Sparse Layered Scene Representation and Adaptive Shading

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Amulet claims that a sparse, tiled, layered cache of potentially visible geometry lets one shaded frame produce many extrapolated frames without neural networks or ghosting.

desk verdict A credible non-neural frame extrapolation pipeline that deserves peer review, but the 'no hallucination' and 250 Hz claims need to be paired with cache coverage bounds and accurate timings. read the letter →

arxiv 2608.10423 v1 pith:5RTPZ5CY submitted 2026-08-11 cs.GR

Sebastian Künzel , Fabian Schmierer , Sergej Geringer , Guido Reina , Daniel Weiskopf , Dieter Schmalstieg (University of Stuttgart , Stuttgart , Germany) This is my paper · ORCID
classification cs.GR
keywords frameextrapolationsparsecachelayeredrenderingdisocclusionadaptiveshadingreusereal-timefroxel
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

Amulet claims that a rendering pipeline can produce many extrapolated display frames from a single fully shaded frame, without neural networks and without the ghosting that motion-vector warping causes. The method replaces the flat framebuffer with a sparse, tiled, layered image-space cache of potentially visible geometry; new views are synthesized by marching rays through this cache front to back, so newly exposed regions are filled from real cached surfaces instead of being hallucinated. Adaptive per-tile reshading, driven by a gradient-based scheduler, keeps shading current where lighting, shadows, or reflections change. If the claim holds, frame extrapolation becomes a non-neural, latency-free way to drive high-refresh displays from modest render rates, and the same cache structure could serve other rendering tasks.

What carries the argument

The central object is a sparse, tiled, layered screen-space cache: each cache is a frustum divided into froxels, with 64 depth layers spaced logarithmically so near layers are thin and far layers are wide. Each froxel maps through a software page table to a tile holding up to one depth-tested visibility sample per pixel location, and only non-empty tiles are allocated. Frame synthesis is a front-to-back ray traversal of this structure with hierarchical empty-space skipping, so most pixels hit cached geometry in the first layers and disoccluded pixels continue deeper. The complementary mechanism is adaptive shading: every tile carries a time to live computed from a 16x-reduced finite-difference shading gradient, spread to neighbours by a max-convolution, and visible tiles whose lifetime expires are reshaded first, while a predicted background cache is shaded in round-robin over multiple frames.

What would settle it

Render a scene with a large disocclusion event—for example, a camera translating sideways past a wall that hides a detailed room—and measure the FLIP error of the first frame after the wall leaves the view; if the extrapolated frame shows holes or missing surfaces where the room should appear, the finite cache did not contain the newly visible geometry, and the central claim fails in exactly that regime.

Watch

Extended reading notes

Core claim

The paper's central claim is that a sparse layered cache is sufficient to extrapolate a shaded frame across several future views at high quality. The cache is built like a k-buffer over froxels (frustum voxels): the view frustum is divided into 64 logarithmically spaced depth layers, each covered by tiles of 16x16 (Full HD) or 32x32 (4K) samples, and only tiles containing geometry are stored. A novel view is produced by casting a ray per pixel through this grid, skipping empty regions with a two-level occupancy mask, and compositing depth-tested samples until the first opaque hit; deeper layers therefore supply correct content where disocclusion would leave a flat G-buffer empty. Shading is amortized by assigning each tile a time to live from finite-difference shading gradients, so tiles whose shading changes rapidly are reshaded immediately while stable tiles persist. The paper reports that in test scenes this approach yields comparable or better visual quality than commercial and neural frame-generation baselines, and can generate nine or more extrapolated frames before error grows linearly.

Load-bearing premise

The load-bearing premise is that the finite layered cache—64 depth layers covering a 25 percent enlarged frustum—already contains every surface that could become visible during the extrapolation interval; if motion exposes geometry that no cache layer captured, the method cannot fill it, and Section 7.7 concedes that non-rigid and animated content must fall back to per-frame shading.

Editorial extensions

If this is right

  • A 60 Hz shading rate can drive a 240 Hz display with no extra latency, because future frames never need to wait for a next keyframe.
  • Disocclusions—regions revealed by camera motion, object motion, or rotation—are filled with actual cached geometry, eliminating the ghosting and blur that flat-frame reprojection produces around thin structures and moving objects.
  • Rasterization, shading, and compositing are decoupled from the display refresh rate, so refresh-rate scaling costs shading work rather than a full re-render per frame.
  • Moving shadows and moving reflections are handled by gradient-triggered tile reshading, so dynamic lighting does not require warping static motion vectors.
  • For non-rigid or heavily deforming content, the method falls back to per-frame updates, so the acceleration shrinks in exactly those scenes (stated limitation, Section 7.7).

Reading between the lines

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

  • The cache is a general potentially-visible representation: the same traversal that synthesizes views could be reused for visibility queries, occlusion culling, or ray-picking, so the paper's contribution may extend beyond frame generation.
  • Because the paper does not combine the cache with spatial upscaling and calls the two orthogonal, a natural next test is stacking both to see whether effective throughput grows multiplicatively.
  • If the linear quality decay after the first extrapolated frames (reported for the 9-frame case) is consistent across scenes, the gradient scheduler could be closed-loop: measure per-frame error slope and automatically adjust tile lifetimes or the cache-swap period.
  • The admitted weakness that small highlights can be missed by the 16x-reduced gradient suggests a testable fix: a second, saliency-weighted gradient pass at full resolution only in tiles containing high-frequency shading, which would preserve the cost savings where gradients are smooth.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper presents Amulet, a real-time frame extrapolation system that replaces the conventional flat framebuffer with a sparse, tiled, layered cache organized into froxels. Visibility samples (depth and primitive ID) and shading are stored in the cache, and novel views are synthesized by ray-marching through the layers; disocclusions are filled from cached layers rather than by motion-vector warping. Dynamic scenes are handled through multiple object caches with warped view rays, and an adaptive shading scheduler assigns per-tile lifetimes based on gradient estimates to amortize shading cost. The method is evaluated on four scenes at Full HD and 4K against DLSS 4.5, MobFGSR, and MoFlow, reporting PSNR, SSIM, LPIPS, and FLIP for quality and GPU timings for performance. The central claim is that Amulet produces high-quality extrapolated frames without hallucination and without neural networks, at performance competitive with state-of-the-art frame generation methods.

Significance. If the claims hold, Amulet would be a significant contribution to real-time rendering: it offers a non-neural, low-latency alternative to neural frame generation, with explicit handling of disocclusions. The paper is strong in its detailed description of the layered cache structure, the multi-cache warping, and the gradient-based shading scheduler. The evaluation against independent ground truth and multiple external baselines, including interpolation methods that have access to future frames, is a notable strength. However, the central 'without hallucination' claim is conditional on the finite cache containing all potentially visible geometry, a condition that is never measured, and the headline performance claim of 250 Hz at 4K is not supported by the reported timings. The quality comparisons also lack error bars, which weakens the quantitative conclusions.

major comments (5)
  1. [Sections 4.3, 5, and 7.7]
  2. [Abstract and Section 7.2, Table 3]
  3. [Tables 5 and 8]
  4. [Section 6.1, Section 7.6, and Figure 12 caption]
  5. [Section 7.7 and Section 7.1]
minor comments (6)
  1. [Section 7.4, Baselines]
  2. [Table 8 caption]
  3. [Section 6.3]
  4. [Section 6.2]
  5. [Figure 12 caption]
  6. [General]

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the paper's quality claims are measured against independent ground truth and external baselines, and its self-cited components are used as implementation substrates rather than as proofs of the central result.

full rationale

Amulet's central claim is that a sparse layered cache can extrapolate multiple frames with accurate disocclusion filling, and this claim is verified empirically: Section 7.4 compares against ground-truth deferred rendering and external baselines (DLSS 4.5, MobFGSR, MoFlow) using PSNR, SSIM, LPIPS, and FLIP, with results reported in Tables 5-7. The main self-cited dependency, the disocclusion buffer of Künzel et al. [2025], is explicitly described as a substrate that Amulet extends: 'Our cache implementation extends the disocclusion buffer introduced by Künzel et al. [2025] with shading data for frame synthesis, multi-object warping and adaptive on-the-fly reshading' (Section 7). The paper does not derive Amulet's quality from that citation; it evaluates the extended system on prerecorded camera paths. The gradient-based scheduler is likewise an algorithmic component whose threshold is hand-set rather than fitted to the reported error metrics; its behavior is tested in dedicated experiments (Section 7.6, Figure 15), not equated to the target results by construction. The 'without hallucination' claim is conditional on the finite 64-layer, 25%-enlarged cache containing all surfaces that become visible during extrapolation, but this is a stated limitation for non-rigid motion (Section 7.7) and a correctness risk for large rigid motion, not a circularity: the paper never defines 'hallucination' in terms of the cache contents, nor does any equation reduce the prediction to the cache's construction. Self-citations to Künzel et al. [2025], Mueller et al. [2021], and Voglreiter et al. [2023] are context and design-choice references and do not carry the burden of the central empirical claim. No fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from prior work to force the design. Overall, the derivation chain is self-contained against independently rendered ground truth and external competitors, so circularity is minimal.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

Amulet's results depend on several hand-chosen parameters: layer count, tile size, amortization period, gradient threshold, gradient resolution, and transparency cap. The undocumented gradient threshold is the largest reproducibility concern. The method relies on standard graphics assumptions such as rigid subcache warping, log-spaced depth layers, and WBOIT. No invented physical entities are introduced.

free parameters (6)
  • Layer count N = 64
    Chosen as a tradeoff between memory and depth quantization error (Section 7.2); sublinear memory scaling shown, but no optimization against quality.
  • Tile size m = 16x16 (Full HD), 32x32 (4K)
    Hand chosen based on page table size and tile fill rate (Section 7); no sensitivity analysis provided.
  • Amortization period delta_t = 4 frames
    Fixed period for background cache generation and shading reuse (Sections 6.2, 7.2); quality and performance depend on it.
  • Gradient threshold = not reported
    Threshold on gradient magnitude used to flag tiles for reshading (Section 6.1); central to adaptive shading quality, but no value is given.
  • Gradient evaluation resolution = 16x reduced resolution
    Initial shading gradients computed at 16x reduced resolution (Section 6.1); the paper admits in Figure 12 caption that this misses a small highlight.
  • Transparency sample cap P = not reported
    Maximum number of semitransparent samples stored per sample location (Section 6.3); affects transparency quality, no value given.
assumptions (4)
  • domain assumption Dynamic objects can be represented with rigid-body transforms
    Subcaches are warped with model matrices only; non-rigid deformation invalidates the cache, as acknowledged in Section 7.7.
  • domain assumption 64 depth layers suffice to represent all relevant scene geometry
    The geometry pass stores up to N samples per pixel; scenes with depth complexity beyond 64 layers will discard samples, and this failure mode is not analyzed.
  • ad hoc to paper Shading changes are predictable from coarse finite differences
    The gradient scheduler uses 16x reduced shading comparisons (Section 6.1); the Figure 12 caption admits that a small highlight was missed, showing the assumption is imperfect.
  • domain assumption Weighted blended order-independent transparency approximates ground-truth blending
    Used for semitransparent compositing (Section 6.3), a standard approximation; authors report no visual artifacts but provide no quantitative validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Amulet: Frame Extrapolation Through Sparse Layered Scene Representation and Adaptive Shading." pith.science (2026). https://pith.science/paper/5RTPZ5CY

@misc{pith2026260810423,
  author       = {Pith},
  title        = {Pith review of: Amulet: Frame Extrapolation Through Sparse Layered Scene Representation and Adaptive Shading},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5RTPZ5CY}},
  note         = {Machine review of arXiv:2608.10423}
}
read the original abstract

We introduce Amulet, a rendering method that transforms a scene into a sparse, tiled and layered intermediate scene representation (cache) for high-frequency frame extrapolation. In contrast to reprojection-based techniques, Amulet explicitly rasterizes and stores potentially visible geometry in its layered image-space cache, allowing accurate shading and inpainting of newly disoccluded regions without hallucination. Our key contribution is a cache that is predictively filled with shading information for future views, amortized over multiple current frames. Novel views are synthesized by hierarchically traversing the cache front to back and refining stale or missing shading on the fly. Using a predictive, gradient-based scheduler that assigns lifetimes for each tile, we enable adaptive shading updates under motion and dynamic lighting. Amulet decouples the rasterization and shading rate from the refresh rate of the display. In many scenarios, our cache can use a single shaded frame to synthesize multiple extrapolated frames with only a few localized updates. In a typical application, we extrapolate a 60 Hz shading rate to a 240 Hz display. Amulet achieves up to 250 Hz at 4K resolution and is competitive with state-of-the-art frame generation methods, including DLSS and neural-flow approaches, in multiple metrics. Amulet explores the design space of sparse layered image-space representation. It enables accurate, non-neural multi frame extrapolation with explicit handling of disocclusions. Our findings show that Amulet can extrapolate many more frames than contemporary methods with high quality, rivaling latency-bound frame interpolation methods with similar quality in many scenes.

Figures

Figures reproduced from arXiv: 2608.10423 by the authors.

Figure 1
Figure 1. The top row shows three frames from the Bistro Exterior scene rendered with our frame extrapolation method: (left) The original keyframe, (middle) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Amulet combines a cache generation stage with a frame extrapolation stage. (left) Two spaceships move toward the position shown in black, as seen from camera position 𝑉0. (middle) Each spaceship is sampled into its own grid cell as indicated by the orange and blue contours. Only grid cells containing at least one sample of an object are stored. These grid cells are highlighted in light blue and orange. The grid cell… view at source ↗
Figure 3
Figure 3. A scene with three objects, A, B, and C. Each object is stored in its [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (11 more)
Figure 5
Figure 5. Figure 5: With our method, the glassware in the Bistro Interior scene can be [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Our test scenes, from top left to bottom right: Bistro Interior, Bistro Exterior, Intel Sponza, and San Miguel. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: The Amulet cache can cover substantial camera movements with extrapolated images. The left and middle images show significant disocclusions on the buildings in the background behind the tree and the street lamp. These frames are extrapolated from the layered cache stru…
Figure 8
Figure 8. Figure 8: Frame extrapolation times (in ms) are shown for an increasing num [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Visual error for a 9x extrapolation case in Bistro Exterior. The plot [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Comparison of Amulet vs. DLSS 4.5 6x multi-frame generation. Shown is the third generated frame after the ground truth by DLSS 4.5 (top), Amulet (middle) and the ground truth (bottom). DLSS 4.5 produces images with visual artifacts around detailed geometry (e.g. visib…
Figure 11
Figure 11. Figure 11: A mirror room with emissive moving objects and multi bounce [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Left: Comparison of the first generated frame (after ground truth) between Amulet, DLSS 4.5 and MobFGSR. The ground truth is shown in the bottom left of the left column. In this view, we focus on artifacts from thin structures, such as window frames. In the overview, …
Figure 13
Figure 13. Figure 13: Left: Comparison of the second generated frame (after ground truth) between Amulet, DLSS 4.5 and MobFGSR. The ground truth is shown in the bottom left of the left column. In this view, we focus on blurring around detailed geometry, such as garden chairs. Amulet and Mo…
Figure 14
Figure 14. Figure 14: Left: Comparison of the first generated frame (after ground truth) between Amulet, MoFlow and MobFGSR. The ground truth is shown in the top right of the left column. In this view, all approaches yield strong extrapolation quality. Right: Comparison of the first genera…
Figure 15
Figure 15. Figure 15: Two columns on the left: Comparison of the second generated frame (after ground truth) between [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 34 canonical work pages

  1. [1]

    Mark and Leonard McMillan and Gary Bishop , title =

    William R. Mark and Leonard McMillan and Gary Bishop , title =. 1997 , numpages =. doi:10.1145/253284.253292 , booktitle =

    work page arXiv 1997

  2. [2]

    ExtraSS: A Framework for Joint Spatial Super Sampling and Frame Extrapolation , year =

    Wu, Songyin and Kim, Sungye and Zeng, Zheng and Vembar, Deepak and Jha, Sangeeta and Kaplanyan, Anton and Yan, Ling-Qi , booktitle =. ExtraSS: A Framework for Joint Spatial Super Sampling and Frame Extrapolation , year =. doi:10.1145/3610548.3618224 , isbn =

    arXiv

  3. [3]

    Layered Depth Images , year =

    Shade, Jonathan and Gortler, Steven and He, Li-wei and Szeliski, Richard , booktitle =. Layered Depth Images , year =. doi:10.1145/280814.280882 , isbn =

    work page arXiv

  4. [4]

    Efficient and Robust From-Point Visibility , year=

    Popescu, Voicu and Sacks, Elisha and Cui, Jian and Ashok, Rohan , journal=. Efficient and Robust From-Point Visibility , year=. doi:10.1109/TVCG.2023.3291138 , numpages =

    work page arXiv 2023

  5. [5]

    The Visual Computer , title =

    Jaroslav Kravec and Martin K. The Visual Computer , title =. 2023 , number =

    work page 2023

  6. [6]

    Computer Graphics Forum , title =

    Hladky, Jozef and Seidel, Hans-Peter and Steinberger, Markus , year =. Computer Graphics Forum , title =. doi:10.1111/cgf.142648 , pages =

    work page doi:10.1111/cgf.142648

  7. [7]

    and Neff, Thomas and Voglreiter, Philip and Steinberger, Markus and Schmalstieg, Dieter , journal =

    Mueller, Joerg H. and Neff, Thomas and Voglreiter, Philip and Steinberger, Markus and Schmalstieg, Dieter , journal =. Temporally Adaptive Shading Reuse for Real-Time Rendering and Virtual Reality , year =. doi:10.1145/3446790 , numpages =

    work page doi:10.1145/3446790

  8. [8]

    2021 , url=

    A Deep Dive into Nanite Virtualized Geometry , author=. 2021 , url=

    work page 2021

Show all 60 references
  1. [9]

    Eurographics Symposium on Rendering , title =

    Baker, Daniel and Jarzynski, Mark , year =. Eurographics Symposium on Rendering , title =

  2. [10]

    and Yang, Jason C

    Hillesland, Karl E. and Yang, Jason C. , pages=. Texel. doi:10.2312/egsh.20161018 , journal =

    doi

  3. [11]

    2002 , booktitle =

    Kraus, Martin and Ertl, Thomas , title =. 2002 , booktitle =

    2002

  4. [12]

    Schaufler, Gernot , booktitle =. Per-. 1998 , editor =

    1998

  5. [13]

    and Nehab, Diego , booktitle =

    Sitthi-amorn, Pitchaya and Lawrence, Jason and Yang, Lei and Sander, Pedro V. and Nehab, Diego , booktitle =. 2008 , editor =. doi:10.2312/EGGH/EGGH08/095-101 , isbn =

    2008 doi

  6. [14]

    and Lefohn, Aaron and Comba, Jo\

    Bavoil, Louis and Callahan, Steven P. and Lefohn, Aaron and Comba, Jo\. Proceedings of the 2007 Symposium on Interactive 3D Graphics and Games , title =. 2007 , address =. doi:10.1145/1230100.1230117 , isbn =

    2007

  7. [15]

    Proceedings of the International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision , pages =

    Diepstraten, Joachim and Weiskopf, Daniel and Kraus, Martin and Ertl, Thomas , title =. Proceedings of the International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision , pages =

  8. [16]

    Computer Graphics Forum , volume =

    Schaufler, Gernot and Stürzlinger, Wolfgang , title =. Computer Graphics Forum , volume =. doi:https://doi.org/10.1111/1467-8659.1530227 , year =

  9. [17]

    Proceedings of SIGGRAPH , title =

    Briedis, Karlis Martins and Djelouah, Abdelaziz and Ortiz, Rapha\". Proceedings of SIGGRAPH , title =. 2023 , address =. doi:10.1145/3588432.3591497 , isbn =

    2023

  10. [18]

    ExtraNet: Real-time Extrapolated Rendering for Low-latency Temporal Supersampling , year =

    Guo, Jie and Fu, Xihao and Lin, Liqiang and Ma, Hengjun and Guo, Yanwen and Liu, Shiqiu and Yan, Ling-Qi , journal =. ExtraNet: Real-time Extrapolated Rendering for Low-latency Temporal Supersampling , year =. doi:10.1145/3478513.3480531 , numpages =

  11. [19]

    Mob-FGSR: Frame Generation and Super Resolution for Mobile Real-Time Rendering , year =

    Yang, Sipeng and Zhu, Qingchuan and Zhuge, Junhao and Qiu, Qiang and Li, Chen and Yan, Yuzhong and Xu, Huihui and Yan, Ling-Qi and Jin, Xiaogang , booktitle =. Mob-FGSR: Frame Generation and Super Resolution for Mobile Real-Time Rendering , year =. doi:10.1145/3641519.3657424 , isbn =

  12. [20]

    , booktitle =

    Torborg, Jay and Kajiya, James T. , booktitle =. Talisman: Commodity Realtime 3D Graphics for the PC , year =. doi:10.1145/237170.237274 , isbn =

  13. [21]

    and Steinberger, Markus and Schmalstieg, Dieter , journal =

    Neff, Thomas and Mueller, Joerg H. and Steinberger, Markus and Schmalstieg, Dieter , journal =. Meshlets and How to Shade Them: A Study on Texture-Space Shading , year =. doi:https://doi.org/10.1111/cgf.14474 , selected =

    doi

  14. [22]

    and Voglreiter, Philip and Dokter, Mark and Neff, Thomas and Makar, Mina and Steinberger, Markus and Schmalstieg, Dieter , journal =

    Mueller, Joerg H. and Voglreiter, Philip and Dokter, Mark and Neff, Thomas and Makar, Mina and Steinberger, Markus and Schmalstieg, Dieter , journal =. Shading Atlas Streaming , year =. doi:10.1145/3272127.3275087 , numpages =

  15. [23]

    Burns and Warren A

    Christopher A. Burns and Warren A. Hunt , journal =. The Visibility Buffer: A Cache-Friendly Approach to Deferred Shading , year =

  16. [24]

    and Winter, Martin and Steinberger, Markus , journal =

    Tatzgern, Wolfgang and Weinrauch, Alexander and Stadlbauer, Pascal and Mueller, Joerg H. and Winter, Martin and Steinberger, Markus , journal =. Radiance Caching with On-Surface Caches for Real-Time Global Illumination , year =. doi:10.1145/3675382 , numpages =

    doi

  17. [25]

    and Moore, Jeremy and Gross, Markus , year =

    Bowles, Huw and Mitchell, Kenny and Sumner, Robert W. and Moore, Jeremy and Gross, Markus , year =. Iterative Image Warping , volume =. doi:10.1111/j.1467-8659.2012.03002.x , journal =

    2012

  18. [26]

    Spatio-Temporal Upsampling on the GPU , year =

    Herzog, Robert and Eisemann, Elmar and Myszkowski, Karol and Seidel, Hans-Peter , booktitle =. Spatio-Temporal Upsampling on the GPU , year =. doi:10.1145/1730804.1730819 , isbn =

  19. [27]

    2018 , isbn =

    Marrs, Adam and Spjut, Josef and Gruen, Holger and Sathe, Rahul and McGuire, Morgan , title =. 2018 , isbn =. doi:10.1145/3231578.3231579 , booktitle =

    2018

  20. [28]

    2016 , publisher =

    Reinert, Bernhard and Kopf, Johannes and Ritschel, Tobias and Cuervo, Eduardo and Chu, David and Seidel, Hans-Peter , title =. 2016 , publisher =

    2016

  21. [29]

    Computer Graphics Forum , year =

    Tessellated Shading Streaming , author=. Computer Graphics Forum , year =. doi:10.1111/cgf.13780 , number =

    doi

  22. [30]

    Proceedings of Computer Graphics and Interactive Techniques , title =

    Andersson, Pontus and Nilsson, Jim and Akenine-M\". Proceedings of Computer Graphics and Interactive Techniques , title =. 2020 , number =. doi:10.1145/3406183 , numpages =

    2020 doi

  23. [31]

    GFFE: G-buffer Free Frame Extrapolation for Low-latency Real-time Rendering , year =

    Wu, Songyin and Vembar, Deepak and Sochenov, Anton and Panneer, Selvakumar and Kim, Sungye and Kaplanyan, Anton and Yan, Ling-Qi , journal =. GFFE: G-buffer Free Frame Extrapolation for Low-latency Real-time Rendering , year =. doi:10.1145/3687923 , issue_date =

    doi

  24. [32]

    ACM Transactions on Graphics , title =

    Ragan-Kelley, Jonathan and Lehtinen, Jaakko and Chen, Jiawen and Doggett, Michael and Durand, Fr\'. ACM Transactions on Graphics , title =. 2011 , issn =. doi:10.1145/1966394.1966396 , issue_date =

    2011

  25. [33]

    Potentially Visible Hidden-Volume Rendering for Multi-View Warping , year =

    Kim, Janghun and Lee, Sungkil , journal =. Potentially Visible Hidden-Volume Rendering for Multi-View Warping , year =. doi:10.1145/3592108 , url =

    doi

  26. [34]

    Trim Regions for Online Computation of From-Region Potentially Visible Sets , year =

    Voglreiter, Philip and Kerbl, Bernhard and Weinrauch, Alexander and Mueller, Joerg Hermann and Neff, Thomas and Steinberger, Markus and Schmalstieg, Dieter , journal =. Trim Regions for Online Computation of From-Region Potentially Visible Sets , year =. doi:10.1145/3592434 , url =

    doi

  27. [35]

    The Camera Offset Space: Real-Time Potentially Visible Set Computations for Streaming Rendering , year =

    Hladky, Jozef and Seidel, Hans-Peter and Steinberger, Markus , journal =. The Camera Offset Space: Real-Time Potentially Visible Set Computations for Streaming Rendering , year =. doi:10.1145/3355089.3356530 , url =

  28. [36]

    and Ortiz-Cayon, Rodrigo and Kalantari, Nima Khademi and Ramamoorthi, Ravi and Ng, Ren and Kar, Abhishek , journal =

    Mildenhall, Ben and Srinivasan, Pratul P. and Ortiz-Cayon, Rodrigo and Kalantari, Nima Khademi and Ramamoorthi, Ravi and Ng, Ren and Kar, Abhishek , journal =. Local Light Field Fusion: Practical View Synthesis with Prescriptive Sampling Guidelines , year =. doi:10.1145/330634...

  29. [37]

    Clarberg, Petrik and Munkberg, Jacob , journal =. Deep. 2014 , number =. doi:10.1145/2661229.2661245 , publisher =

    2014

  30. [38]

    , journal =

    Tole, Parag and Pellacini, Fabio and Walter, Bruce and Greenberg, Donald P. , journal =. Interactive Global Illumination in Dynamic Scenes , year =. doi:10.1145/566654.566613 , publisher =

  31. [39]

    Proceedings of SIGGRAPH Asia , publisher =

    Potentially Visible Set Generation with the Disocclusion Buffer , author=. Proceedings of SIGGRAPH Asia , publisher =. 2025 , series =. doi:10.1145/3757377.3763981 , articleno =

    2025

  32. [40]

    Game Developers Conference , year=

    FidelityFX Super Resolution 2.0 , author=. Game Developers Conference , year=

  33. [41]

    Journal of Computer Graphics Techniques , volume =

    Weighted Blended Order-Independent Transparency , author=. Journal of Computer Graphics Techniques , volume =

  34. [42]

    Proceedings of VRST , publisher =

    Impostor-based Rendering Acceleration for Virtual, Augmented, and Mixed Reality , author=. Proceedings of VRST , publisher =. doi:10.1145/3489849.3489865 , articleno =

  35. [43]

    Computer Graphics Forum , volume=

    Reshadable Impostors with Level-of-Detail for Real-Time Distant Objects Rendering , author=. Computer Graphics Forum , volume=. 2025 , doi=

    2025

  36. [44]

    and Lawrence, Jason and Nehab, Diego and Hoppe, Hugues and Wilkins, Clara L

    Yang, Lei and Tse, Yu-Chiu and Sander, Pedro V. and Lawrence, Jason and Nehab, Diego and Hoppe, Hugues and Wilkins, Clara L. , title =. 2011 , volume =. doi:10.1145/2070781.2024184 , journal =

    2011

  37. [45]

    ACM Transactions on Graphics , volume=

    MoFlow: Motion-Guided Flows for Recurrent Rendered Frame Prediction , author=. ACM Transactions on Graphics , volume=. 2025 , publisher=. doi:10.1145/3730400 , articleno =

    2025 doi

  38. [46]

    Proceedings of I3D , pages=

    Decoupled Deferred Shading for Hardware Rasterization , author=. Proceedings of I3D , pages=. doi:10.1145/2159616.2159640 , year=

  39. [47]

    2018 , pages=

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric , author=. 2018 , pages=. doi:10.1109/CVPR.2018.00068 , booktitle=

    2018

  40. [48]

    IEEE Transactions on Image Processing , volume=

    Image quality assessment: from error visibility to structural similarity , author=. IEEE Transactions on Image Processing , volume=. doi:10.1109/TIP.2003.819861 , year=

    2003

  41. [49]

    Kallweit, Simon and Clarberg, Petrik and Kolb, Craig and Davidovi. The. 2022 , month =

    2022

  42. [50]

    2018 , url=

    Ben Lang , title=. 2018 , url=

    2018

  43. [51]

    ACM Transactions on Graphics , volume =

    Wu, Zhizhen and Cao, Zhe and Yuan, Yazhen and Yuan, Zhilong and Wang, Rui and Huo, Yuchi , title =. ACM Transactions on Graphics , volume =. 2025 , doi =

    2025

  44. [52]

    Proceedings of SIGGRAPH Asia , numpages =

    Zuo, Chenyu and Yuan, Yazhen and Wu, Zhizhen and Liu, Zhijian and Lan, Jingzhen and Fu, Ming and Huo, Yuchi and Wang, Rui , title =. Proceedings of SIGGRAPH Asia , numpages =. 2025 , doi =

    2025

  45. [53]

    Computer Graphics Forum , volume =

    Vining, Nicholas and Majercik, Alexander and Gu, Floria and Takikawa, Towaki and Trusty, Ty and Lalonde, Paul and McGuire, Morgan and Sheffer, Alla , title =. Computer Graphics Forum , volume =. doi:https://doi.org/10.1111/cgf.70010 , year =

    doi

  46. [54]

    Learning from

    Evans, Alex , month = aug, year =. Learning from

  47. [55]

    2022 , volume =

    Hladky, Jozef and Stengel, Michael and Vining, Nicholas and Kerbl, Bernhard and Seidel, Hans-Peter and Steinberger, Markus , title =. 2022 , volume =. doi:10.1145/3550454.3555524 , journal =

    2022

  48. [56]

    2025 , volume =

    Lu, Edward and Rowe, Anthony , title =. 2025 , volume =. doi:10.1145/3731213 , journal =

    2025 doi

  49. [57]

    Scher , title =

    Bishop, Gary and Fuchs, Henry and McMillan, Leonard and Zagier, Ellen J. Scher , title =. 1994 , isbn =. doi:10.1145/192161.192195 , booktitle =

    1994

  50. [58]

    2018 , publisher =

    Franke, Linus and Hofmann, Nikolai and Stamminger, Marc and Selgrad, Kai , title =. 2018 , publisher =. doi:10.1145/3203200 , journal =

    2018 doi

  51. [59]

    Computer Graphics Forum , volume =

    Zeng, Zheng and Xu, Zilin and Wang, Lu and Wu, Lifan and Yan, Ling-Qi , title =. Computer Graphics Forum , volume =. doi:10.1111/cgf.14882 , year =

    doi

  52. [60]

    2025 , volume =

    Yang, Sipeng and Ji, Jiayu and Zhuge, Junhao and Zhao, Jinzhe and Qiu, Qiang and Li, Chen and Yan, Yuzhong and Wang, Kerong and Yan, Lingqi and Jin, Xiaogang , title =. 2025 , volume =. doi:10.1145/3763348 , journal =

    2025 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.