Writing study: InfiniBench: Infinite Benchmarking for Visual Spatial Reasoning with Customizable Scene Complexity

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Haoming Wang, Qiyao Xue, Wei Gao · Official CVPR 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: Datasets, benchmarks & measurement (primary).

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

12 physical pages

Before references / appendix

≈ 8.11 page-equivalents

Main figures / tables

13 / 6

Attached appendix figures / tables

0 / 0

Attached appendix span

≈ 3.11 page-equivalents

Before references / appendix: ≈ 67.58%References / intervening material: ≈ 6.5%Attached appendix: ≈ 25.92%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	≈ 1.34 pp.	11.17%
2. Related Work and Motivation	3	≈ 0.97 pp.	8.08%
3. Method	4	≈ 1.52 pp.	12.67%
4. Iteration: The camera moves to the new viewpoint, the	5	≈ 1.94 pp.	16.17%
5. Using InfiniBench to Diagnose VLMs’ Fail-	7	≈ 1.21 pp.	10.08%
6. Conclusion	8	≈ 0.14 pp.	1.17%
Acknowledgments	9	≈ 0.06 pp.	0.5%
References	9	≈ 0.78 pp.	6.5%
A. Vedaldi. Syncity: Training-free generation of 3d	9	≈ 0.55 pp.	4.58%
B. Chen. Camera keyframing with style and control.	10	≈ 1.63 pp.	13.58%
C. Li. Video instruction tuning with synthetic data,	12	≈ 0.93 pp.	7.75%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 3
• Figure 4 · p. 4
• Figure 5 · p. 4
• Figure 6 · p. 5
• Figure 7 · p. 5
• Figure 8 · p. 5
• Table 1 · p. 6
• Table 2 · p. 6
• Figure 9 · p. 7
• Table 3 · p. 7
• Table 4 · p. 7
• Table 5 · p. 8
• Figure 10 · p. 8
• Figure 11 · p. 8
• Figure 12 · p. 8
• Figure 13 · p. 8
• Table 6 · p. 8

Attached appendix

No recognized caption labels.

Separate source files—excluded from canonical-PDF totals

A separate author version may repeat main text and references. Whole-file lengths and counts are not extra appendix material and are never added to the canonical PDF totals. Read the study's source scope for which sections received substantive reading versus layout inspection.

Additional source 1

Separate PDF (or ZIP containing it): whole-file layout of 9 physical pages; 12 detected figures and 2 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 8.87 page-equivalents, 12 figure labels and 2 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
A. In-Context Learning for Scene Constraint	1	≈ 1.45 pp.
B. CoT Reasoning for Refinement of Scene	2	≈ 0.22 pp.
C. Details of Cluster-Based Optimization	2	≈ 0.59 pp.
D. Gallery of Generated 3D Scenes	3	≈ 0.21 pp.
E. Details of the Camera Trajectory Optimiza-	3	≈ 2.25 pp.
F. Details of InfiniBench Implementation	5	≈ 1.88 pp.
G. Additional Qualitative Comparisons	7	≈ 0.63 pp.
H. Mitigating the Impact of Ambiguity in Ob-	8	≈ 0.47 pp.
References	8	≈ 1.17 pp.

• Figure 1 · p. 3
• Figure 2 · p. 4
• Figure 3 · p. 4
• Figure 4 · p. 5
• Figure 5 · p. 5
• Figure 6 · p. 6
• Figure 7 · p. 6
• Figure 8 · p. 7
• Figure 9 · p. 7
• Figure 10 · p. 7
• Figure 11 · p. 8
• Table 1 · p. 8
• Table s · p. 9
• Figure 12 · p. 9

Argument and organization

1. Method §§3.1–3.3

   Translate prompts to constraints, optimize layout and choose informative camera trajectories.

2. Evaluation §§4–5

   Validate the generator itself before using its controlled scenes to diagnose VLMs.

Problem, claim and contribution

Instead of asking for another fixed dataset, the paper identifies lack of control over compositional, relational and observational complexity as the evaluation gap.

Abstract and introduction, PDF pp. 1–2 · Main paper · Additional source 2

Argument and section order

Generator correctness and downstream model diagnosis occupy separate main sections, avoiding the circular claim that difficult questions alone establish a good generator.

§4, PDF pp. 5–7; §5, PDF pp. 7–8 · Main paper · Additional source 2

Figures, captions and visual explanation

The motivation page makes two failure taxonomies concrete: invalid layout geometry and uninformative viewpoints. The labels say what is wrong with each scene rather than leaving failures to visual inference.

Figures 2–3, PDF p. 3 · Main paper · Additional source 2

Tables, comparisons and evidence

Two parallel tables vary object count and occupancy while reporting prompt fidelity, image alignment, realism and physical violations separately; the shared asset library is named in the setup.

Tables 1–2 and §4.1, PDF p. 6 · Main paper · Additional source 2

Appendix and reproducibility

The supplement exposes constraint-refinement examples, layout/camera details, rendering controls, camera pseudocode and an ambiguity check distinguishing object identification from spatial measurement.

Supplement §§A–H, PDF pp. 1–9, Algorithm 1 and Table 1 · Main paper · Additional source 2

Clarity and writing conventions

The downstream study uses named complexity axes and separates clutter, irrelevant objects and viewpoint changes, letting each result answer a controlled question.

§§5.1–5.3, PDF p. 8; Tables 5–6 and Figures 10–13 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For a benchmark generator, first prove that control parameters generate valid intended scenes, then vary those parameters to investigate model failure modes.

Boundary: Theoretically unlimited procedural variants are not unlimited real-world diversity; model-scored realism and a bounded asset library constrain the evidence.

Apply this lesson using the argument and evidence worksheet.

Inspection record and sources

Checked

2026-09-26

Reading status

Writing-focused manuscript scan + selected visuals inspected

Main text

Manuscript-wide writing scan of PDF pp. 1–12: framing, argument sequence, section organization, figure/table captions, evaluation interpretation and limitations; reference sections identified. This is a writing-focused scan, not an exhaustive technical proof or reproduction audit. Main PDF pages 3 and 6 were rendered and visually inspected.

Appendix / supplement

Writing-focused scan of the separate supplementary PDF, pp. 1–9, including its organization, captions, protocol/evidence detail and limitation material where present; proofs and code listings were inspected for exposition, not independently verified or executed. Supplement §§A–H, PDF pp. 1–9, Algorithm 1 and Table 1. Linked video/audio demonstrations and non-PDF archive contents were not assessed.

Visually inspected PDF pages

3, 6

This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.

• Official CVPR 2026 CVF open-access accepted manuscript; retrieved 2026-09-26 · 12 pages
• Official CVPR 2026 CVF supplementary PDF; retrieved 2026-09-26 · 9 pages