GATE: Geometry-Aware Trained Encoding

Carsten Benthin; Daniel Meister; Jakub Bok\v{s}ansk\'y

REVIEW 4 major objections 4 minor 28 references

GATE: Geometry-Aware Trained Encoding

T0 review · 4 major / 4 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 This paper proposes GATE, a positional encoding that stores learnable feature vectors directly on triangle surfaces and interpolates them with barycentric coordinates, and reports that in neural ambient occlusion and neural radiance…

desk verdict GATE is a genuine geometry-aware encoding with real promise, but the headline speedups are inflated by a sparse-update confound that needs a baseline. read the letter →

arxiv 2506.08161 v2 pith:MIV3CVBC submitted 2025-06-09 cs.GR

Jakub Bokšanský , Daniel Meister , Carsten Benthin This is my paper

classification cs.GR

keywords geometry-awaretrainedencodingmeshcolorsbarycentricinterpolationtrainablefeaturevectorsneuralradiancecachingambientocclusionmulti-resolutionhashonlinerendering

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

GATE is a positional encoding for neural networks that process points on 3D meshes: instead of looking up feature vectors in a volumetric grid or hash table, it places learnable feature vectors on triangle surfaces and interpolates the three closest ones with barycentric coordinates. The paper argues that this aligns encoding density with actual scene geometry, removes the need to normalize positions to scene bounds, avoids hash collisions, and gives linear, cache-friendly memory access. Applying GATE to neural ambient occlusion and neural radiance caching, the paper reports up to 3.7x faster training and up to 2.7x faster inference than a multi-resolution hash-grid encoding at comparable memory, with equal or lower image error as measured by the FLIP metric. The largest gains appear on scenes where a small detailed object sits inside a large environment. If the claims hold, online-trained neural rendering becomes cheaper and more robust to extreme scale variation.

What carries the argument

The load-bearing object is mesh colors, a scheme that virtually tessellates a triangle into smaller triangles at a chosen resolution $R$ and places feature vectors at all vertices of the virtual tessellation, giving $\frac{(R+1)(R+2)}{2}$ feature vectors per triangle. A query point is encoded by taking the three feature vectors at the vertices of the virtual triangle that contains it and computing their barycentric interpolation; the virtual-triangle indices and weights come from the integral and fractional parts of the barycentric coordinates scaled by $R$. Around this, the method builds a linear feature-vector buffer with per-mesh offsets, an adaptive resolution rule $R_I = \mathrm{clamp}(32\,A_I^2\,R_{\mathrm{scale}}, 1, 32)$ based on the average normalized triangle area, stacked resolutions whose feature vectors are concatenated, per-triangle training counters that adjust the stochastic optimizer's step-count parameters and bias sample selection toward rarely trained triangles, and a sparse-gradient update that only touches feature vectors used by the current training batch.

What would settle it

Run the same neural ambient occlusion and neural radiance caching training with a hash-grid that records and updates only the feature vectors touched by the current training batch, matching GATE's sparse-update behavior, and compare per-frame training time and FLIP error; if that sparse hash-grid matches or beats GATE's time at equal quality, then the speed advantage comes from sparse updating rather than geometry-aware placement.

Watch

Extended reading notes

Core claim

The central claim is that the right place to store trainable features for surface-based neural rendering is on the surface itself. GATE distributes feature vectors over each triangle using the mesh-colors scheme at a resolution $R$, so a triangle carries $\frac{(R+1)(R+2)}{2}$ feature vectors; a query point on the triangle is encoded by locating the three closest feature vectors in the virtual tessellation and interpolating them with barycentric weights. Because the query is tied to a triangle, the encoding is implicitly normalized and its density follows the geometry, and an adaptive per-triangle resolution sets $R$ from the triangle's world-space area to avoid wasting memory on tiny triangles. The paper evaluates GATE in neural ambient occlusion and neural radiance caching, and reports up to 3.7x faster training and up to 2.7x faster inference than multi-resolution hash-grid encoding at comparable memory, with FLIP error that is usually lower and never substantially higher; the largest advantages occur when a detailed object is surrounded by a large scene.

Load-bearing premise

The speed comparison in Section 4 assumes a hash-grid baseline that, as the paper's own footnote states, trains every feature vector in the grid each step, while GATE instead updates only the feature vectors touched by the current training samples, so the reported speedups mix the encoding design with a sparse-update trick.

Editorial extensions

If this is right

Because GATE's feature density is set by triangle size, users no longer need to tune grid resolution against scene size or worry about hash collisions for surface queries.
Scenes with extreme scale variation, such as a detailed teapot inside a stadium, can be encoded without giving the whole scene volume high resolution.
Training cost per frame scales with the number of triangles touched by the training samples, so large scenes become cheaper to train than with a hash grid whose every entry is updated each step.
Per-triangle training metadata enables sample distribution and learning-rate scheduling tailored to how often each surface region has been trained, which the paper uses to improve convergence quality.
With comparable memory, GATE is reported to match or improve FLIP error in most tested scenes, so the encoding does not trade quality for speed.

Reading between the lines

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

The reported training speedups in Section 4 conflate geometry-aware placement with sparse updates, since the hash-grid baseline follows the reference implementation that updates all feature vectors each step; a sparse-update hash-grid comparison would isolate the encoding's true contribution.
Because the encoding is tied to triangle IDs and barycentric coordinates, the same machinery could be dropped into any ray-tracing pipeline that already knows which triangle a ray hit, not just ambient occlusion and radiance caching.
The paper's observation that GATE does not benefit from normal and albedo inputs hints that the triangle identity itself carries much of the surface information, which suggests the encoding could serve other surface-attribute learning tasks.
The mesh-colors basis is stressed by elongated triangles, so replacing it with a per-face texture parametrization would extend GATE to production meshes with uneven tessellations.

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.

Desk Editor's Note

read the letter

Colleague,

GATE is a real, usable encoding: storing trainable feature vectors on triangle surfaces with barycentric interpolation, using mesh colors to distribute them. That is a sensible and well-executed idea for surface-based neural rendering, and it avoids hash-grid problems like collisions and scene-bound normalization. The adaptive per-triangle resolution and per-triangle training counter/learning rate are genuinely new combinations, and the ablation on adaptive resolution is convincing. The writing is clear and honest about limitations, e.g., mesh colors on elongated triangles, which helps.

The soft spot is the central speed claim. The abstract and Table 1 say up to 3.7x / 2.7x speedups, but the footnote admits the hash-grid baseline trains every feature vector every step while GATE only updates touched ones. That is a confound: any parametric encoding could do sparse updates, so the speedup is not attributable to the geometry-aware encoding itself. The inference speedups (1.7-3x) are more credible because inference does not have that sparsity advantage, but even those could be affected by implementation quality. A sparse-update hash-grid baseline is necessary before claiming an encoding advantage.

The quality claim is also slightly overstated. Section 4 says 'for all example scenes' GATE delivers higher quality, but Figure 5 itself says it fails to match hash-grid on Kitchen, and Table 1 shows Bathroom NRC FLIP higher (0.414 vs 0.409). That is not a big deal—the ratios are close—but the universal wording should be softened to 'equal or better in most scenes.'

Minor: no error bars, R_scale is a per-scene tuned parameter, and the per-triangle learning rate is another implementation difference that is not isolated. None of these are fatal, but they mean the paper's headline numbers should be read as upper bounds.

Bottom line: the method is solid and worth engaging with. The paper deserves a serious referee round, but the authors should add a sparse-update hash-grid baseline, report variance across runs, and temper the claims. I would accept it for peer review with that expectation, and I would cite it once the speed claims are cleaned up.

Referee Report

4 major / 4 minor

Summary. The manuscript introduces GATE, a trainable positional encoding in which latent feature vectors are stored on the surface of triangular meshes via mesh colors and interpolated with barycentric coordinates. It includes an adaptive per-mesh resolution rule, a sparse feature-vector update strategy, and per-triangle Adam bookkeeping, and it evaluates the encoding for neural ambient occlusion and neural radiance caching. The reported results claim training speedups up to 3.7x and inference speedups up to 2.7x while matching or improving FLIP error relative to a multi-resolution hash-grid baseline.

Significance. The geometry-aligned feature layout is a plausible and potentially useful alternative to volume hash grids for online neural rendering; removing scene-bounds normalization and addressing the teapot-in-the-stadium scenario are genuine advantages. However, the experimental section as written does not isolate the contribution of the encoding from implementation-level and input-dimensionality differences, so the headline quantitative claims are not yet supported. The adaptive resolution idea is interesting but depends on a per-scene free parameter that is not reported.

major comments (4)

[Section 3.2 and Section 4 (Table 1 footnote)] The training-time comparison is not an apples-to-apples test of the encoding. GATE updates only the feature vectors touched by the current training batch, while the hash-grid baseline follows the reference implementation and updates every feature vector in the grid each step, as the footnote acknowledges. Any parametric encoding can adopt the sparse-update strategy, so the reported speedups (up to 56.2x in Table 2) conflate a general optimizer optimization with the proposed geometry-aware layout. A sparse-update hash-grid baseline, with the same number of updated parameters per step, is required before 'faster training' can be attributed to GATE.
[Section 4, Figure 5, and Table 1] The text states 'For all example scenes, GATE is able to deliver a higher image quality', but this is contradicted by Figure 5's caption (Kitchen NRC is an exception) and by Table 1's Bathroom NRC row, where GATE's FLIP is 0.414 versus 0.409 for hash-grid. The universal 'higher quality' claim must be withdrawn or qualified, and the Bathroom case should be discussed.
[Section 4 (MLP setup)] The inference speed comparison is confounded by different network input sizes and input sets. GATE uses 4 input neurons for NAO and 16 for NRC, whereas the 8-level hash-grid with 4 features per level produces 32 input values, and for NRC the hash-grid additionally consumes normal and albedo while GATE does not. The 1.7-3x inference speedup may therefore be due to a smaller decoder rather than the claimed cache-friendly memory access. A matched-input-dimension comparison or per-component timing is needed.
[Equation (2) and Section 4] R_scale is a scene-specific, hand-tuned hyperparameter, but the paper does not report the values used for each scene or any sensitivity analysis. Because this parameter directly controls feature-vector density and therefore the memory/quality trade-off, the reported quality comparisons and memory ratios are not reproducible without it.

minor comments (4)

[Abstract and Table 2] The abstract and conclusion say the training speedup is up to 3.7x, while Table 2 reports up to 56.2x for the large hash-grid; please state which comparison is being summarized.
[Tables 1 and 2, Figures 4-6] No error bars or repeated-run statistics are reported for FLIP or timing; given the small differences in some rows (e.g., 0.021 vs 0.021 and 0.033 vs 0.034 in Table 1), a few repeated runs would strengthen the quality claims.
[Section 5] The paper acknowledges that mesh colors handle elongated triangles poorly and introduce memory overhead for finely tessellated meshes; these limitations should be surfaced in Section 4's discussion of scene selection rather than deferred to future work.
[Equation (3)] In Equation (3), the middle term appears to have a typo: it should presumably be ∂z/∂z_t rather than ∂z/z_t.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GATE's claims are empirical comparisons with independent baselines, and no derivation reduces to its own inputs.

full rationale

The paper does not claim a first-principles derivation of its encoding; GATE is a proposed geometry-aware encoding evaluated empirically against multi-resolution hash-grid encoding. The only scene-specific parameter, R_scale, adjusts feature-vector density as a clamped quadratic function of triangle area (Eq. 2), and it does not encode the target error metric, so the quality results are not self-fulfilling. The use of mesh colors is attributed to prior external work [YKH10] and is not a self-citation, and no uniqueness theorem or ansatz is imported from the authors' own prior publications. The headline speedups compare GATE against a reference hash-grid implementation that updates every grid feature each step (Table 1 footnote), while GATE updates only the feature vectors touched by the current training batch (Section 3.2). This is a legitimate baseline-fairness or attribution concern about whether the speed advantage comes from the encoding or from sparse updates, but it is not circularity: the comparison is not definitionally forced, and the paper does not present the sparse-update strategy as a derived prediction. No equation in the paper reduces to another by construction, and no fitted parameter is renamed as a prediction. Therefore the manuscript contains no significant circularity.

Assumptions & free parameters 8 free parameters · 8 assumptions · 0 invented entities

No new physical entities are introduced. The method is an engineering contribution: it transfers mesh colors distribution to trainable feature vectors and adds per-triangle training heuristics. The central claim is empirical, not derived, so the ledger records the hyperparameters and domain assumptions that the reported numbers depend on.

free parameters (8)

R_scale = not stated, tuned per scene
Scene-specific density parameter in Eq. 2 (R_I = clamp(32 * A_I^2 * R_scale, 1, 32)); controls feature vector density per mesh and affects memory and quality.
mesh colors resolution R = R=1 for fixed level; adaptive R_I via Eq. 2
Hyperparameter determining number of feature vectors per triangle: (R+1)(R+2)/2. GATE stacks one adaptive and one fixed R=1 resolution.
feature vector length L = 2 per resolution for NAO, 8 per resolution for NRC
Length of each latent feature vector (Section 3, between 1 and 8); with two stacked resolutions gives 4 and 16 input neurons.
number of stacked resolutions = 2
Authors use two mesh colors resolutions hierarchically (Section 4); number is a hyperparameter.
training sample count per iteration = 49152
Set in Section 4; one training iteration per frame with batch size equal to sample count.
candidate samples M = 16
M=16 random samples per iteration in Eq. 4 for sample distribution.
training count cap = 512
Cap in the weight w(x_Ti) = 1 / min(512, t_Ti) from Section 3.2.1.
hash-grid baseline configuration = 8 levels, 4 features/level, base resolution 2; L/M/S sizes
Comparison baseline parameters from Section 4; selected to give comparable memory (L) or smaller (M,S). Speedups are relative to this configuration.

assumptions (8)

standard math Barycentric coordinate interpolation (Eq. 1) yields a valid encoding of surface points on a triangle.
Standard linear interpolation; loaded feature vectors from the three closest vertices of a virtual triangle.
standard math Distributing feature vectors via mesh colors [YKH10] covers triangle surfaces at a chosen resolution R.
The paper reuses the mesh colors algorithm to place feature vectors at virtual triangle vertices; no new derivation.
domain assumption Input points to the MLP always lie on triangle surfaces (except camera rays), so per-triangle encoding suffices.
Stated in Section 1: rays originate from triangle surfaces, so GATE only needs surface positions.
domain assumption Triangle geometry and mesh connectivity are static during online training.
Training happens over a fixed mesh; dynamic scenes would need continuous retraining of the encoding, which the paper mentions but does not evaluate.
domain assumption Sharing feature vectors at vertices and edges via the index buffer guarantees seamless encoding across shared triangle edges.
Section 3.1 says this holds with fixed R per mesh.
ad hoc to paper Per-triangle learning rate and sample distribution heuristics improve convergence.
Section 3.2.1 introduces these mechanisms without theoretical analysis; they are claimed to improve training quality and speed.
ad hoc to paper Adaptive resolution formula R_I = clamp(32 * A_I^2 * R_scale, 1, 32) provides a good trade-off between accuracy and memory.
Empirical heuristic in Eq. 2; no optimality guarantee and R_scale is scene-specific.
domain assumption A small MLP (2 hidden layers, 32 neurons, leaky ReLU) is a sufficient decoder for the encoded features.
Used for all experiments (Section 4); not validated across larger/smaller networks.

how reviews work

0 comments

Cite this review

Pith. "Pith review of GATE: Geometry-Aware Trained Encoding." pith.science (2026). https://pith.science/paper/MIV3CVBC

@misc{pith2026250608161,
  author       = {Pith},
  title        = {Pith review of: GATE: Geometry-Aware Trained Encoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MIV3CVBC}},
  note         = {Machine review of arXiv:2506.08161}
}

read the original abstract

The encoding of input parameters is one of the fundamental building blocks of neural network algorithms. Its goal is to map the input data to a higher-dimensional space, typically supported by trained feature vectors. The mapping is crucial for the efficiency and approximation quality of neural networks. We propose a novel geometry-aware encoding called GATE that stores feature vectors on the surface of triangular meshes. Our encoding is suitable for neural rendering-related algorithms, for example, neural radiance caching. It also avoids limitations of previous hash-based encoding schemes, such as hash collisions, selection of resolution versus scene size, and divergent memory access. Our approach decouples feature vector density from geometry density using mesh colors, while allowing for finer control over neural network training and adaptive level-of-detail.

Figures

Figures reproduced from arXiv: 2506.08161 by the authors.

**Figure 1.** Bathroom scene (top) rendered with neural radiance caching (NRC) and Bistro-Exterior (bottom) rendered with neural ambient occlusion (NAO) after 1024 (NRC) and 128 (NAO) training steps using multi-dimensional hash-grid encoding [MESK22] (left), our GATE method (middle), and ground truth (right). GATE provides faster training (up to 3.7×) and inference times (up to 2.7×) and at the same time achieves higher quality (… view at source ↗

**Figure 2.** An overview of our method. Latent feature vectors are evenly distributed on triangles using mesh colors [YKH10] with resolution R. We find a virtual triangle formed by vertices (vi ,vj ,vk) where the input point x belongs to. We use barycentric coordinates within this virtual triangle (wi ,wj ,wk) to interpolate of three feature vectors (zi , z j , zk) associated with the vertices to get the encoded vector z = (z1,.… view at source ↗

**Figure 3.** An example of mesh colors [YKH10] that we use to distribute feature vectors on triangles using different resolutions. The feature vectors from multiple resolutions can be concatenated to improve the approximation quality. fed to the MLP in a similar manner as for multi-resolutions hashgrids, where the number of resolutions is a hyperparameter. 3.2. Training We train the feature vectors together with the MLP using … view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: Neural ambient occlusion (NAO). Image quality after 128 training iterations: hash-grid (left), GATE (middle), and ground truth (right). Size of hash-grid < 1MB. GATE provides higher image quality (lower FLIP error) than hash-grid, in particular for the two larger scene…

**Figure 5.** Figure 5: Neural radiance caching (NRC). Image quality after 1024 training iterations: hash-grid (left), GATE (middle), and ground truth (right). Size of hash-grid < 1 MB. GATE is able to match or exceed the image quality of hash-grid, except for the Kitchen scene. Even though G…

**Figure 6.** Figure 6: A comparison of the multi-resolution hash-grid encoding (top) and our GATE (bottom) for the Bistro (Ext) scene rendered with the neural ambient occlusion (NAO) with 1 sample per pixel, showing how the FLIP error decreases with an increasing number of training steps (up…

**Figure 7.** Figure 7: A comparison of GATE with a fixed resolution (left) and the adaptive one (right) in the Kitchen scene. Voronoi diagram (bottom) visualizes the closest feature vector location, using the same color for points belonging to the same feature vector. Notice how the adaptiv…

**Figure 8.** Figure 8: A comparison of NRC + GATE after 1024 training iterations with (left) and without (right) additional G-Buffer inputs such as normal and albedo. Compared to hash-grids, GATE does not benefit from these two additional inputs. With additional inputs, the neural network u…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 25 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry.original add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 i...
[3]

: Visualizing and Communicating Errors in Rendered Images

Andersson P., Nilsson J., Akenine - M \" o ller T. : Visualizing and Communicating Errors in Rendered Images . In Ray Tracing Gems II, Marrs A., Shirley P., Wald I., (Eds.). 2021, ch. 19, pp. 301--320

work page 2021
[4]

: Ptex: Per-face texture mapping for production rendering

Burley B., Lacewell D. : Ptex: Per-face texture mapping for production rendering. Computer Graphics Forum 27, 4 (2008), 1155--1164

work page 2008
[5]

: Neural two-level monte carlo real-time rendering

Dereviannykh M., Klepikov D., Hanika J., Dachsbacher C. : Neural two-level monte carlo real-time rendering. Computer Graphics Forum (2025)

work page 2025
[6]

Neural Texture Block Compression

Fujieda S., Harada T. : Neural texture block compression. arXiv preprint arXiv:2407.09543 (2024)

work page Pith review arXiv 2024
[7]

: Local Positional Encoding for Multi-Layer Perceptrons

Fujieda S., Yoshimura A., Harada T. : Local Positional Encoding for Multi-Layer Perceptrons . In Pacific Graphics Short Papers and Posters (2023), Chaine R., Deng Z., Kim M. H., (Eds.), The Eurographics Association

work page 2023
[8]

M., Tagliasacchi A

Govindarajan S., Sambugaro Z., Shabhanov A., Takikawa T., Sun Weiweiand Rebain D., Conci N., Yi K. M., Tagliasacchi A. : Lagrangian hashing for compressed neural field representations. In ECCV (2024)

work page 2024

Show all 28 references

[9]

: Neural radiosity

Hadadan S., Chen S., Zwicker M. : Neural radiosity. ACM Trans. Graph. 40, 6 (Dec. 2021)

2021
[10]

: Adam: A method for stochastic optimization

Kingma D., Ba J. : Adam: A method for stochastic optimization. International Conference on Learning Representations (12 2014)

2014
[11]

: Neumip: multi-resolution neural materials

Kuznetsov A., Mullia K., Xu Z., Ha s an M., Ramamoorthi R. : Neumip: multi-resolution neural materials. ACM Trans. Graph. 40, 4 (July 2021)

2021
[12]

: Instant neural graphics primitives with a multiresolution hash encoding

M \"u ller T., Evans A., Schied C., Keller A. : Instant neural graphics primitives with a multiresolution hash encoding. ACM transactions on graphics (TOG) 41, 4 (2022), 1--15

2022
[13]

Martel J. N. P., Lindell D. B., Lin C. Z., Chan E. R., Monteiro M., Wetzstein G. : Acorn: adaptive coordinate networks for neural scene representation. ACM Trans. Graph. 40, 4 (July 2021)

2021
[14]

: Neural importance sampling

M \"u ller T., McWilliams B., Rousselle F., Gross M., Nov \'a k J. : Neural importance sampling. ACM Transactions on Graphics (ToG) 38, 5 (2019), 1--19

2019
[15]

: Real-time neural radiance caching for path tracing

M\" u ller T., Rousselle F., Nov\' a k J., Keller A. : Real-time neural radiance caching for path tracing. ACM Trans. Graph. 40, 4 (July 2021)

2021
[16]

P., Tancik M., Barron J

Mildenhall B., Srinivasan P. P., Tancik M., Barron J. T., Ramamoorthi R., Ng R. : Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV (2020)

2020
[17]

: On the spectral bias of neural networks

Rahaman N., Baratin A., Arpit D., Draxler F., Lin M., Hamprecht F., Bengio Y., Courville A. : On the spectral bias of neural networks. In Proceedings of the 36th International Conference on Machine Learning (09--15 Jun 2019), Chaudhuri K., Salakhutdinov R., (Eds.), vol. 97 of ...

2019
[18]

: Neural geometry fields for meshes

Sivaram V., Li T.-M., Ramamoorthi R. : Neural geometry fields for meshes. In ACM SIGGRAPH 2024 Conference Papers (New York, NY, USA, 2024), SIGGRAPH '24, Association for Computing Machinery

2024
[19]

: Variable bitrate neural fields

Takikawa T., Evans A., Tremblay J., M\" u ller T., McGuire M., Jacobson A., Fidler S. : Variable bitrate neural fields. In ACM SIGGRAPH 2022 Conference Proceedings (New York, NY, USA, 2022), SIGGRAPH '22, Association for Computing Machinery

2022
[20]

: Neural geometric level of detail: Real-time rendering with implicit 3D shapes

Takikawa T., Litalien J., Yin K., Kreis K., Loop C., Nowrouzezahrai D., Jacobson A., McGuire M., Fidler S. : Neural geometric level of detail: Real-time rendering with implicit 3D shapes
[21]

: Compact neural graphics primitives with learned hash probing

Takikawa T., M\" u ller T., Nimier-David M., Evans A., Fidler S., Jacobson A., Keller A. : Compact neural graphics primitives with learned hash probing. In SIGGRAPH Asia 2023 Conference Papers (2023)

2023
[22]

: Deferred neural rendering: image synthesis using neural textures

Thies J., Zollh\" o fer M., Nie ner M. : Deferred neural rendering: image synthesis using neural textures. ACM Trans. Graph. 38, 4 (July 2019)

2019
[23]

N., Kaiser L

Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A. N., Kaiser L. u., Polosukhin I. : Attention is all you need. In Advances in Neural Information Processing Systems (2017), Guyon I., Luxburg U. V., Bengio S., Wallach H., Fergus R., Vishwanathan S., Garnett R.,...

2017
[24]

: Random-access neural compression of material textures

Vaidyanathan K., Salvi M., Wronski B., Akenine-M \"o ller T., Ebelin P., Lefohn A. : Random-access neural compression of material textures. arXiv preprint arXiv:2305.17105 (2023)

2023 arXiv
[25]

: Neural prefiltering for correlation-aware levels of detail

Weier P., Zirr T., Kaplanyan A., Yan L.-Q., Slusallek P. : Neural prefiltering for correlation-aware levels of detail. ACM Trans. Graph. 42, 4 (July 2023)

2023
[26]

Yuksel C., Keyser J., House D. H. : Mesh colors. ACM Transactions on Graphics (TOG) 29, 2 (2010), 1--11

2010
[27]

: Rethinking texture mapping

Yuksel C., Lefebvre S., Tarini M. : Rethinking texture mapping. Computer Graphics Forum 38, 2 (2019), 535--551

2019
[28]

: Real-time neural appearance models

Zeltner* T., Rousselle* F., Weidlich* A., Clarberg* P., Nov \'a k* J., Bitterli* B., Evans A., Davidovi c T., Kallweit S., Lefohn A. : Real-time neural appearance models. ACM Transactions on Graphics 43, 3 (2024), 1--17

2024

Pith tools

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

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry.original add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 i...

[3] [3]

: Visualizing and Communicating Errors in Rendered Images

Andersson P., Nilsson J., Akenine - M \" o ller T. : Visualizing and Communicating Errors in Rendered Images . In Ray Tracing Gems II, Marrs A., Shirley P., Wald I., (Eds.). 2021, ch. 19, pp. 301--320

work page 2021

[4] [4]

: Ptex: Per-face texture mapping for production rendering

Burley B., Lacewell D. : Ptex: Per-face texture mapping for production rendering. Computer Graphics Forum 27, 4 (2008), 1155--1164

work page 2008

[5] [5]

: Neural two-level monte carlo real-time rendering

Dereviannykh M., Klepikov D., Hanika J., Dachsbacher C. : Neural two-level monte carlo real-time rendering. Computer Graphics Forum (2025)

work page 2025

[6] [6]

Neural Texture Block Compression

Fujieda S., Harada T. : Neural texture block compression. arXiv preprint arXiv:2407.09543 (2024)

work page Pith review arXiv 2024

[7] [7]

: Local Positional Encoding for Multi-Layer Perceptrons

Fujieda S., Yoshimura A., Harada T. : Local Positional Encoding for Multi-Layer Perceptrons . In Pacific Graphics Short Papers and Posters (2023), Chaine R., Deng Z., Kim M. H., (Eds.), The Eurographics Association

work page 2023

[8] [8]

M., Tagliasacchi A

Govindarajan S., Sambugaro Z., Shabhanov A., Takikawa T., Sun Weiweiand Rebain D., Conci N., Yi K. M., Tagliasacchi A. : Lagrangian hashing for compressed neural field representations. In ECCV (2024)

work page 2024