Writing study: SAM 3D: 3Dfy Anything in Images

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Xingyu Chen, Fu-Jen Chu, Pierre Gleize, Kevin J Liang, Alexander Sax, Hao Tang, Weiyao Wang, Michelle Guo, Thibaut Hardin, Xiang Li, Aohan Lin, Jia-Wei Liu, Ziqi Ma, Anushka Sagar, Bowen Song, Xiaodong Wang, Jianing "Jed" Yang, Bowen Zhang, Piotr Dollár, Georgia Gkioxari, Matt Feiszli, Jitendra Malik · 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

13 physical pages

Before references / appendix

≈ 8.04 page-equivalents

Main figures / tables

9 / 4

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 61.85%References / intervening material: ≈ 38.15%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	≈ 0.99 pp.	7.62%
2. The SAM 3D Model	2	≈ 0.92 pp.	7.08%
3. Training SAM 3D	3	≈ 3.56 pp.	27.38%
4. Experiments	7	≈ 1.13 pp.	8.69%
5. Related Work	8	≈ 0.28 pp.	2.15%
6. Conclusion	8	≈ 0.13 pp.	1%
Acknowledgements	8	≈ 0.25 pp.	1.92%
References	9	≈ 4.96 pp.	38.15%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 3 · p. 4
• Figure 4 · p. 4
• Table 1 · p. 4
• Figure 5 · p. 5
• Figure 6 · p. 6
• Table 2 · p. 6
• Table 3 · p. 7
• Figure 7 · p. 7
• Figure 8 · p. 7
• Figure 9 · p. 8
• Table 4 · 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 19 physical pages; 12 detected figures and 8 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 18.94 page-equivalents, 12 figure labels and 8 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.22 pp.
A. Related Work	1	≈ 0.85 pp.
B. Data Annotation Engine Details	2	≈ 4.41 pp.
C. Pretraining and Mid-Training Data Details	6	≈ 1.52 pp.
D. Details on Model Training	8	≈ 3.51 pp.
E. Evaluation	11	≈ 2.21 pp.
F. Additional Ablations	13	≈ 2.14 pp.
G. Potential Negative Societal Impacts	15	≈ 0.74 pp.
H. Limitations	16	≈ 3.34 pp.

• Figure 10 · p. 3
• Figure 11 · p. 3
• Figure 12 · p. 4
• Figure 13 · p. 5
• Figure 14 · p. 5
• Figure 16 · p. 10
• Table 5 · p. 11
• Table 6 · p. 11
• Figure 15 · p. 11
• Table 7 · p. 13
• Table 8 · p. 13
• Table 9 · p. 14
• Table 10 · p. 14
• Table 11 · p. 15
• Table 12 · p. 15
• Figure 17 · p. 15
• Figure 18 · p. 16
• Figure 19 · p. 16
• Figure 20 · p. 18
• Figure 21 · p. 19

Argument and organization

1. Problem and contribution

   SAM 3D frames realistic single-image reconstruction as a data problem as well as an architecture problem. The main text devotes substantial space to how useful annotations and preferences are obtained, making the data engine a visible contribution.

2. Argument sequence

   Architecture is followed by pretraining, mid-training and post-training before evaluation/analysis; related work comes later. The order keeps the long data-generation story adjacent to the model it trains rather than interrupting it with a catalog of predecessors.

3. Evidence and verification

   The evaluation treats shape, layout and texture separately. The texture preference comparison holds geometry fixed, while the data-engine progression and cumulative-training analysis connect human annotation effort to model changes.

Problem, claim and contribution

SAM 3D frames realistic single-image reconstruction as a data problem as well as an architecture problem. The main text devotes substantial space to how useful annotations and preferences are obtained, making the data engine a visible contribution.

Introduction; Section 3 · Main paper · Additional source 2

Argument and section order

Architecture is followed by pretraining, mid-training and post-training before evaluation/analysis; related work comes later. The order keeps the long data-generation story adjacent to the model it trains rather than interrupting it with a catalog of predecessors.

Sections 2–6 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 separates geometry/layout from texture/refinement and enlarges the attention-sharing pattern. Figure 3 contrasts four data families, and Figure 4 plus Table 1 align stages with sources and conditioning inputs; these answer different questions instead of repeating one pipeline.

Figures 2–4 and Table 1, PDF pp.3–4 · Main paper · Additional source 2

Tables, comparisons and evidence

The evaluation treats shape, layout and texture separately. The texture preference comparison holds geometry fixed, while the data-engine progression and cumulative-training analysis connect human annotation effort to model changes.

Section 4; Figure 8 and Table 3, PDF p.7; Figure 9 and Table 4, PDF p.8 · Main paper · Additional source 2

Appendix and reproducibility

The long supplement documents annotation interfaces, synthetic rendering, training objectives/hyperparameters, evaluation, further ablations, impacts and limitations. This modular organization allows a reader to inspect either data labor or learning details without searching an undifferentiated appendix.

Supplementary Sections B–H, PDF pp.1–19 · Main paper · Additional source 2

Clarity and writing conventions

The training narrative labels pre-, mid- and post-training explicitly and explains the meaning of preference/Elo comparisons. Such labels distinguish measured human preference from geometric error and prevent one score from becoming a universal quality claim.

Section 3; preference and data-engine analyses in Section 4 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For data-intensive foundation models, prepare architecture, data-family and annotation-process figures separately; isolate shape, layout and texture when evaluating outputs.

Boundary: Preference win rates depend on the candidate set and study protocol. Cumulative training stages show a recipe's trajectory but do not by themselves isolate every interaction causally.

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

Writing-oriented scan throughout main PDF pp.1–8 and references pp.9–13: abstract/framing, section sequence, figure/table captions, evaluation organization and conclusions; not an exhaustive technical verification.

Appendix / supplement

Writing-oriented supplement scan: all 19 supplementary PDF pages scanned for data-engine, rendering, objective, evaluation, ablation, impact and limitation organization. Video/code assets were not executed or assessed.

Visually inspected PDF pages

3, 4

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 CVF open-access CVPR 2026 accepted-paper PDF · 13 pages
• Official CVF supplementary PDF · 19 pages