Writing study: Volume Transformer: Revisiting Vanilla Transformers for 3D Scene Understanding
← ECCV 2026 oral writing corpus
Kadir Yilmaz, Adrian Kruse, Tristan Höfer, Daan de Geus, Bastain Leibe · Official ECCV 2026 oral record
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
- 20 physical pages
- Before references / appendix
- ≈ 15.15 page-equivalents
- Main figures / tables
- 4 / 4
- 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.
| Top-level heading | Starts on PDF page | Estimated span | Share of entire PDF |
|---|---|---|---|
| 1 Introduction | 1 | ≈ 2.59 pp. | 12.95% |
| 2 Related Work | 4 | ≈ 1.25 pp. | 6.25% |
| 3 Method | 5 | ≈ 2.98 pp. | 14.9% |
| 4 Training Volt | 8 | ≈ 2.78 pp. | 13.9% |
| 5 Experiments | 11 | ≈ 4.14 pp. | 20.7% |
| 6 Conclusion | 15 | ≈ 0.25 pp. | 1.25% |
| Acknowledgements | 15 | ≈ 0.43 pp. | 2.15% |
| References | 16 | ≈ 4.85 pp. | 24.25% |
All detected figure and table locations
Main text
Attached appendix
No recognized caption labels.
Argument and organization
- 1 Introduction; 2 Related Work
Motivates architectural simplicity through compatibility with the broader Transformer ecosystem, while immediately acknowledging data scarcity and overfitting in 3D.
- 3 Method; 4 Training Volt
Separates architecture from training recipe, making it possible to distinguish the value of a simple backbone from the distillation and data needed to train it.
- 5 Experiments; 6 Conclusion
Outdoor results separate single-dataset from jointly trained models, list parameter counts and preserve validation/test columns rather than pooling incompatible training regimes.
Problem, claim and contribution
Motivates architectural simplicity through compatibility with the broader Transformer ecosystem, while immediately acknowledging data scarcity and overfitting in 3D.
Abstract and Section 1, pp.1–3 · Main paper
Argument and section order
Separates architecture from training recipe, making it possible to distinguish the value of a simple backbone from the distillation and data needed to train it.
Sections 3–4, pp.5–10 · Main paper
Figures, captions and visual explanation
The architecture keeps the volumetric patch input, linear tokenizer, global encoder and light decoder visually distinct; the correspondence to per-point output is explicit.
Figure 2, p.6 · Main paper
Tables, comparisons and evidence
Outdoor results separate single-dataset from jointly trained models, list parameter counts and preserve validation/test columns rather than pooling incompatible training regimes.
Table 2, p.13 · Main paper
Appendix and reproducibility
The retrieved manuscript contains the main argument and reference pages; no separate supplemental source was retrieved, so supplemental implementation coverage is not asserted.
Main PDF pp.1–20; official oral event 6170 · Main paper
Clarity and writing conventions
A dedicated training section prevents the simplicity claim from hiding a substantive training recipe, and the patch-size ablation acknowledges the accuracy/latency trade-off.
Section 4, pp.8–10; Table 4, p.15 · Main paper
What to reuse—and what not to infer
Transferable lesson: If claiming a simpler architecture, isolate architectural changes from teacher supervision, optimization and added data; report training regimes and hardware alongside efficiency.
Boundary: A vanilla-style encoder is not a no-adaptation or no-distillation result. Joint training and single-dataset rows cannot be compared as if supervision were identical.
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
- 5, 6, 13
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.