Writing study: SPARK: Sim-ready Part-level Articulated Reconstruction with VLM Knowledge

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Yumeng He, Ying Jiang, Jiayin Lu, Yin Yang, Chenfanfu Jiang · 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

11 physical pages

Before references / appendix

≈ 8.13 page-equivalents

Main figures / tables

7 / 4

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 73.91%References / intervening material: ≈ 26.09%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	≈ 0.56 pp.	5.09%
2. Related Work	2	≈ 0.91 pp.	8.27%
3. Method	3	≈ 2.62 pp.	23.82%
4. Experiments	5	≈ 2.95 pp.	26.82%
5. Conclusion	8	≈ 0.17 pp.	1.55%
Acknowledgments	9	≈ 0.09 pp.	0.82%
References	9	≈ 2.87 pp.	26.09%

All detected figure and table locations

Main text

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

Located appendix only: starts on physical p.1; ≈ 5.88 page-equivalents, 2 figure labels and 1 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
Supplementary Material	1	≈ 0.04 pp.
6. VLM Ablation	1	≈ 0.17 pp.
7. Joint Origin Optimization Results	1	≈ 0.39 pp.
8. Network Architecture Details	1	≈ 5.28 pp.

• Table 5 · p. 1
• Figure 8 · p. 1
• Figure 9 · p. 1

Argument and organization

1. Problem and contribution

   SPARK defines a simulation-ready result as more than a visually plausible mesh: part geometry and articulated joint parameters are both required. The introduction uses this gap to motivate combining semantic priors with geometric generation/refinement.

2. Argument sequence

   The method is partitioned into VLM priors, part-level generation, joint optimization and data construction. Evaluation mirrors that decomposition with distinct shape and URDF measurements, then component ablations and downstream examples.

3. Evidence and verification

   Table 1 evaluates shape using Chamfer distance and two F-score thresholds; Table 2 separately measures joint axis, pivot and type errors. The text identifies 100 images covering 25 categories, preserving the scope behind the aggregate scores.

Problem, claim and contribution

SPARK defines a simulation-ready result as more than a visually plausible mesh: part geometry and articulated joint parameters are both required. The introduction uses this gap to motivate combining semantic priors with geometric generation/refinement.

Introduction; Section 3 · Main paper · Additional source 2

Argument and section order

The method is partitioned into VLM priors, part-level generation, joint optimization and data construction. Evaluation mirrors that decomposition with distinct shape and URDF measurements, then component ablations and downstream examples.

Sections 3.1–3.4 and 4 · Main paper · Additional source 2

Figures, captions and visual explanation

Figure 2 shows VLM-generated reference parts, open-state images and URDF proposals feeding different branches, followed by local/global/hierarchical attention and differentiable joint refinement. Rendered parts and articulated states make the output contract visible.

Figure 2, PDF p.4 · Main paper · Additional source 2

Tables, comparisons and evidence

Table 1 evaluates shape using Chamfer distance and two F-score thresholds; Table 2 separately measures joint axis, pivot and type errors. The text identifies 100 images covering 25 categories, preserving the scope behind the aggregate scores.

Tables 1–2 and Section 4.1, PDF p.7 · Main paper · Additional source 2

Appendix and reproducibility

The supplement documents VLM alternatives, joint-origin optimization, diffusion/conditioning details, part merging and watertight/articulation preparation. Those are important dependencies of the apparent end-to-end pipeline, not incidental implementation trivia.

Supplementary PDF pp.1–6 · Main paper · Additional source 2

Clarity and writing conventions

The in-the-wild caption states that noisy or non-frontal inputs may first be converted into generated front views. Naming that preprocessing makes a qualitative gallery interpretable and avoids implying that every output was conditioned identically.

Figure 5 caption, PDF p.7 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For compositional generation, align method subsections with separately measurable outputs: part geometry, joint parameters and downstream behavior each need their own evidence.

Boundary: VLM-generated descriptions and views are proposals, not ground truth. A robot-manipulation demonstration does not establish general reliability across unseen articulation types.

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.9–11: abstract/framing, section sequence, figure/table captions, evaluation organization and conclusions; not an exhaustive technical verification.

Appendix / supplement

Writing-oriented supplement scan: all six supplementary PDF pages, with VLM/joint studies, diffusion implementation, part processing and articulation-data preparation. Video/code assets were not executed or assessed.

Visually inspected PDF pages

4, 7

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 · 11 pages
• Official CVF supplementary ZIP member PDF; see URL fragment for member · 6 pages