Writing study: Make Geometry Matter for Spatial Reasoning
← ECCV 2026 oral writing corpus
Shihua Zhang, Qiuhong Shen, Shizun Wang, Tianbo Pan, Xinchao Wang · Official ECCV 2026 oral record
Figure, table and section ledger
Editorial contribution type: Training, objectives & optimization (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
- 19 physical pages
- Before references / appendix
- ≈ 14.79 page-equivalents
- Main figures / tables
- 3 / 7
- Attached appendix figures / tables
- 0 / 0
- Attached appendix span
- No boundary detected; not assumed zero
Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.
Source-bound correction: All top-level main headings and the reference boundary checked against source PDF text; source-bound positions use physical single-column y/page-height, not a two-column reading-order surrogate. Running page-number/author titles, affiliations, numbered contribution bullets, footnotes and bibliography continuations are excluded. This linked manuscript contains no appended supplementary section; separate supplement links were not scanned.
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.
All detected figure and table locations
Main text
Attached appendix
No recognized caption labels.
Argument and organization
- 1 Introduction; 2 Related Work
Starts with the negative observation that adding geometry can have little benefit or even hurt, rather than assuming extra tokens must be useful.
- 3 Method: geometry-aware baseline, masking and guided fusion
Defines the existing geometry-injection framework before marking the two new strategies; static and dynamic settings remain separate throughout evaluation.
- 4 Experiments and analysis; 5 Conclusion
Training ablations test the modules, while inference-time removal or zeroing of geometry tests whether the trained model relies on that input.
Problem, claim and contribution
Starts with the negative observation that adding geometry can have little benefit or even hurt, rather than assuming extra tokens must be useful.
Figure 1 and Section 1, pp.1–3 · Main paper
Argument and section order
Defines the existing geometry-injection framework before marking the two new strategies; static and dynamic settings remain separate throughout evaluation.
Sections 3–4, pp.5–14 · Main paper
Figures, captions and visual explanation
The proposed diagram separates training-time masking from fusion and shows the conditioning paths; the baseline diagram is explicitly described as the starting framework.
Figure 2, p.6; Figure 3, p.7 · Main paper
Tables, comparisons and evidence
Training ablations test the modules, while inference-time removal or zeroing of geometry tests whether the trained model relies on that input.
Tables 3–5, pp.13–14 · Main paper
Appendix and reproducibility
Further geometry-reliance and overhead analyses are pointed to in the appendix; the scan assessed those pointers but not a separately retrieved appendix.
Sections 4.3–4.4, pp.14–15 · Main paper
Clarity and writing conventions
The wording distinguishes geometric information being present from being used, and the tables repeat static/dynamic labels instead of hiding them in a pooled average.
Section 3.2, pp.7–8; Table 5, p.14 · Main paper
What to reuse—and what not to infer
Transferable lesson: To claim a model uses a new modality, test the modality at inference, not only the loss at training, and preserve a no-modality baseline.
Boundary: A larger removal drop demonstrates dependence under that intervention, not by itself a causal guarantee of better reasoning in every setting.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-02
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-oriented scan across the manuscript main argument, physical PDF pp.1–15: framing and section roadmaps, method/argument explanations, figure and table captions, comparison protocols, conclusions and limitations. Selected full-page rasters were inspected. This is not exhaustive line-by-line technical reading, proof verification or experimental reproduction.
- Appendix / supplement
- Main-manuscript pointers to appendix/supplement and the reference boundary were checked. Separate supplementary PDFs were not retrieved or inspected; their contents are not included in inspected-page counts.
- Visually inspected PDF pages
- 7, 14
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.