Writing study: CoSMo3D: Open-World Promptable 3D Semantic Segmentation through LLM-Guided Canonical Spatial Modeling

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Li Jin, Weikai Chen, Yujie Wang, Yingda Yin, Zeyu HU, Runze Zhang, Keyang Luo, Shengju Qian, Xin Wang, Xueying Qin · 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: Architecture & representation (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

10 physical pages

Before references / appendix

≈ 8.09 page-equivalents

Main figures / tables

6 / 2

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 80.9%References / intervening material: ≈ 19.1%Estimated layout envelopes, not text/artwork area. Unresolved appendix boundaries remain within the undivided post-body material.

Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.

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	≈ 0.86 pp.	8.6%
2. Related Work	2	≈ 1.02 pp.	10.2%
3. Method	3	≈ 2.55 pp.	25.5%
4. Experiment	6	≈ 2.54 pp.	25.4%
5. Conclusions	8	≈ 0.18 pp.	1.8%
Acknowledgments	9	≈ 0.04 pp.	0.4%
References	9	≈ 1.91 pp.	19.1%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 4
• Figure 4 · p. 5
• Table 1 · p. 7
• Figure 5 · p. 7
• Table 2 · p. 8
• Figure 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 4 physical pages; 5 detected figures and 3 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 3.93 page-equivalents, 5 figure labels and 3 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.02 pp.
1. Outline	1	≈ 0.07 pp.
2. More Canonicalization Details	1	≈ 0.57 pp.
3. fluid air port	1	≈ 0 pp.
4. light output port	1	≈ 0.01 pp.
5. weapon attack part	1	≈ 0 pp.
6. manual tool part	1	≈ 0.01 pp.
7. electronic operate part	1	≈ 0 pp.
8. instrument play part	1	≈ 0.01 pp.
9. general container part	1	≈ 0 pp.
10. plant container part	1	≈ 0.01 pp.
11. bath container part	1	≈ 0 pp.
12. wearable fit part	1	≈ 0.01 pp.
13. rotate axis part	1	≈ 0 pp.
14. arch fixed part	1	≈ 0.01 pp.
15. measure display part	1	≈ 0 pp.
16. special function part	1	≈ 0.01 pp.
17. furniture support part	1	≈ 0 pp.
18. transport carry part	1	≈ 0.01 pp.
19. heat electric part	1	≈ 1.46 pp.
References	3	≈ 1.77 pp.

• Table S1 · p. 1
• Table S2 · p. 1
• Table S3 · p. 1
• Figure S1 · p. 2
• Figure S2 · p. 2
• Figure S3 · p. 2
• Figure S4 · p. 2
• Figure S5 · p. 4

Argument and organization

1. Canonical-space motivation

   Explain why pose-coordinate semantics fail before introducing cross-category canonical data.

2. Data and model as distinct contributions

   Separate canonicalization from the dual-branch objective, then test pose, prompts and module variants.

Problem, claim and contribution

Functional part examples such as a supporting leg turn canonicalization into a semantic problem rather than only an alignment operation.

Introduction, PDF pp. 1–2 · Main paper · Additional source 2

Argument and section order

The method overview explicitly separates external data/supervision from the internal architecture/objectives, preventing two sources of gain from being conflated.

Section 3.1, PDF p. 3 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 labels the canonical embedding branch as training-only and uses a divider between feature extraction and supervision. Frozen/trainable icons and distinct loss paths clarify what survives at inference.

Figure 2, PDF p. 3; visual inspection · Main paper · Additional source 2

Tables, comparisons and evidence

The evaluation varies both pose state and prompt form, while baseline retraining on the constructed data is marked in the comparison caption.

Section 4.2 and Table 1, PDF pp. 6–7 · Main paper · Additional source 2

Appendix and reproducibility

The supplement exposes GPT clustering prompts and the resulting semantic groups, then provides a component-by-component visual ablation matching the main variant labels.

Supplement Tables S1–S3 and Figures S1–S4, PDF pp. 1–2 · Main paper · Additional source 2

Clarity and writing conventions

The implementation and evaluation subsections name dataset, point sampling, prompts and rotated/canonical settings before interpreting gains.

Section 4, PDF p. 6 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: When both curated supervision and a model change contribute, draw their roles separately and retrain a baseline on the same data before attributing all improvement to architecture.

Boundary: Open-world and arbitrary-pose wording should remain tied to the evaluated prompt and object distributions, not read as an exhaustive guarantee.

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. 1–10: 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. 1–4, 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 Tables S1–S3 and Figures S1–S4, PDF pp. 1–2. 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 · 10 pages
• Official CVPR 2026 CVF supplementary PDF; retrieved 2026-09-26 · 4 pages