Writing study: Radiometrically Consistent Gaussian Surfels for Inverse Rendering

ICLR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← ICLR 2026 oral writing corpus

Kyu Beom Han, Jaeyoon Kim, Woo Jae Kim, Jinhwan Seo, Sung-eui Yoon · Official ICLR 2026 oral record

Observed writing choices, not acceptance rules. This study examines the inspected manuscript version. It does not establish why the paper was accepted or that its presentation choices should be copied unchanged.

Figure, table and section ledger

Editorial contribution type: Learning & optimization algorithms (primary) · Architecture & representation design.

Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.

Canonical PDF

33 physical pages

Before references / appendix

≈ 9.46 page-equivalents

Main figures / tables

6 / 2

Attached appendix figures / tables

14 / 10

Attached appendix span

≈ 19.89 page-equivalents

Before references / appendix: ≈ 28.67%References / intervening material: ≈ 11.06%Attached appendix: ≈ 60.27%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Section-by-section lengths and source pages

The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.

Top-level heading	Starts on PDF page	Estimated span	Share of entire PDF
1 INTRODUCTION	1	≈ 1.45 pp.	4.39%
2 RELATED WORKS	3	≈ 0.6 pp.	1.82%
3 PRELIMINARIES	3	≈ 0.9 pp.	2.73%
4 METHOD	4	≈ 2.95 pp.	8.94%
5 EXPERIMENTS	7	≈ 2.55 pp.	7.73%
6 CONCLUSION AND FUTURE WORKS	10	≈ 0.21 pp.	0.64%
ACKNOWLEDGEMENTS	10	≈ 0.15 pp.	0.45%
REFERENCES	10	≈ 3.64 pp.	11.03%
APPENDIX	14	≈ 0.11 pp.	0.33%
A IMPLEMENTATION DETAILS	14	≈ 2.17 pp.	6.58%
B COMAPARISON ON RELIGHTING PERFORMANCE AND RENDERING COST	16	≈ 0.29 pp.	0.88%
C VISUAL COMPARISON ON ILLUMINATION COMPONENTS	16	≈ 2.44 pp.	7.39%
D ADDITIONAL VISUAL COMPARISON ON BENCHMARK DATASETS	19	≈ 4 pp.	12.12%
E EVALUATION OF INDIRECT ILLUMINATION	23	≈ 3 pp.	9.09%
F ABLATION STUDIES ON RADIOMETRIC CONSISTENCY	26	≈ 2 pp.	6.06%
G ABLATION STUDY ON HYPERPARAMETERS Ng AND Ns	28	≈ 2 pp.	6.06%
H ADDITIONAL ANALYSIS ON THE INITIALIZATION STRATEGY	30	≈ 1 pp.	3.03%
I ENVIRONMENT MAP RECONSTRUCTION ON BENCHMARK DATASETS	31	≈ 1 pp.	3.03%
J ADDITIONAL ANALYSIS ON FINETUNING-BASED RELIGHTING STRATEGY	32	≈ 1.89 pp.	5.73%

All detected figure and table locations

Main text

• Figure 1 · p. 2
• Figure 2 · p. 5
• Table 1 · p. 7
• Figure 3 · p. 8
• Figure 4 · p. 8
• Table 2 · p. 8
• Figure 5 · p. 9
• Figure 6 · p. 9

Attached appendix

• Table 3 · p. 16
• Figure 7 · p. 17
• Figure 8 · p. 18
• Figure 9 · p. 19
• Figure 10 · p. 20
• Figure 11 · p. 21
• Figure 12 · p. 22
• Table 4 · p. 23
• Figure 13 · p. 24
• Figure 14 · p. 25
• Table 5 · p. 26
• Table 6 · p. 27
• Table 7 · p. 28
• Table 8 · p. 28
• Table 9 · p. 29
• Table 10 · p. 29
• Figure 15 · p. 29
• Table 11 · p. 30
• Figure 16 · p. 30
• Figure 17 · p. 31
• Figure 18 · p. 31
• Figure 19 · p. 32
• Figure 20 · p. 32
• Table 12 · p. 33

Argument and organization

1. Unobserved-direction failure → physical consistency

   Identifies the missing supervision before introducing a new residual objective.

2. Principle → ray tracing → two-stage framework

   Separates the conceptual constraint from its efficient implementation.

3. Quality → ablation → adaptation cost

   Supports both the physical explanation and the relighting speed/quality trade-off.

Problem, claim and contribution

The gap is precise: radiance trained only from camera views remains unconstrained along directions needed for indirect illumination. Related work differentiates cached radiance, learnable residual light and differentiable tracing.

Sections 1–2, pp. 1–3 · Main paper

Argument and section order

The residual is defined before Monte Carlo sampling, initialization losses and relighting; this prevents the long loss inventory from obscuring the actual new idea.

Sections 4.1–4.2, pp. 4–7 · Main paper

Figures, captions and visual explanation

Figure 2 combines a training flowchart with a small ray geometry showing why an occluded surfel receives no camera supervision. Forward/backward line styles connect the physical residual to the optimization graph.

Figure 2, p. 5 · Main paper

Tables, comparisons and evidence

Ablations detach each side of the residual and remove the initialization constraint. Appendix E adds explicitly rendered indirect-light ground truth when standard benchmarks cannot directly test the claimed mechanism.

Figure 6, p. 9; Appendix E, Table 4, p. 23 · Main paper

Appendix and reproducibility

Separate tables distinguish ray-traced PBR, PBR with adapted radiance and direct adapted-surface rasterization, alongside memory and per-frame cost; the upfront adaptation time remains a separate quantity.

Appendix B, Table 3, p. 16; Appendix J, Table 12, p. 33 · Main paper

Clarity and writing conventions

The conclusion limits the material model to dielectric surfaces. Qualitative grids preserve method columns and scene-property rows, while captions identify the particular illumination error to inspect.

Section 6, p. 10; Figures 3 and 5, pp. 8–9 · Main paper

What to reuse—and what not to infer

Transferable lesson: When proposing a physical consistency loss, visualize the missing supervision and create a direct measurement of the hidden quantity rather than relying only on final image quality.

Boundary: Sub-10-ms rendering follows per-light adaptation, not immediate arbitrary relighting. The paper has inconsistent default sampled-surfel counts (4096 in Section 5.1 versus 2048 in Appendix G); implementation reproduction should resolve this instead of silently selecting one.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-19

Reading status

Full text + selected visuals inspected

Main text

Pages 1–10, Sections 1–6: complete formulation, implementation, benchmark comparisons and ablations.

Appendix / supplement

Pages 14–33, Appendices A–J: implementation, relighting, component visualizations, new indirect-light evaluation, sampling and initialization studies.

Visually inspected PDF pages

5, 16, 23, 33

• ICLR 2026 official proceedings PDF, retrieved 2026-09-19 · 33 pages