Writing study: Featurising Pixels from Dynamic 3D Scenes with Linear In-Context Learners

CVPR overviewCorpus atlasBy contribution typeVisual galleryRules & templateWriting patternsPaper studies

← CVPR 2026 oral writing corpus

Nikita Araslanov, Martin Sundermeyer, Hidenobu Matsuki, David Joseph Tan, Federico Tombari · 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: 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

10 physical pages

Before references / appendix

≈ 8.05 page-equivalents

Main figures / tables

8 / 5

Attached appendix figures / tables

0 / 0

Attached appendix span

Separate file recorded below; not in this PDF total

Before references / appendix: ≈ 80.5%References / intervening material: ≈ 19.5%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	≈ 0.95 pp.	9.5%
2. Related work	2	≈ 0.59 pp.	5.9%
3. Linear In-Context Learning	2	≈ 1.77 pp.	17.7%
4. Experiments	4	≈ 4.26 pp.	42.6%
5. Conclusion	8	≈ 0.16 pp.	1.6%
References	9	≈ 1.95 pp.	19.5%

All detected figure and table locations

Main text

• Figure 1 · p. 1
• Figure 2 · p. 3
• Figure 2b · p. 3
• Figure 3 · p. 4
• Table 1 · p. 5
• Figure 4 · p. 5
• Table 2 · p. 6
• Figure 5 · p. 6
• Table 3 · p. 7
• Table 4 · p. 8
• Table 5 · 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 2 physical pages; 0 detected figures and 1 tables. These may include overlapping main-text material.

Located appendix only: starts on physical p.1; ≈ 1.69 page-equivalents, 0 figure labels and 0 table labels. This interval is distinct from the whole file.

Detected heading	PDF page	Estimated span
A. Qualitative examples	1	≈ 0.16 pp.
B. Backbone generalisation	1	≈ 0.16 pp.
C. Practical details	1	≈ 0.01 pp.
D. Zero-shot linear probing	1	≈ 0.73 pp.
References	2	≈ 0.62 pp.

• Table 6 · p. 1

Argument and organization

1. Linear in-context learning §3

   Define context/query pairs, cue construction and the shared linear-map objective.

2. Experiments §§4.1–4.4

   Probe video segmentation, surface normals and semantics, then vary cues, training components and temporal gap.

Problem, claim and contribution

The gap combines spatial resolution and temporal consistency: image-pretrained patch features and video-level action representations each leave a missing dense capability.

Abstract and §1, PDF pp. 1–2 · Main paper · Additional source 2

Argument and section order

The method explains what remains fixed across frames—the context-fitted linear map—before describing the decoder and noisy teacher cues, making the invariance test understandable.

§3, PDF pp. 2–4 · Main paper · Additional source 2

Figures, captions and visual explanation

The training overview puts the context and query frames on a time axis, labels the shared map and shows depth/flow/feature cue production in a separate inset.

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

Tables, comparisons and evidence

Table 2 groups results by backbone and lists training data; the adjacent Figure 5 pairs quantitative normal/semantic probes with ground truth, feature projections and predictions for the same examples.

Table 2 and Figure 5, PDF p. 6 · Main paper · Additional source 2

Appendix and reproducibility

A compact supplement expands backbone generalization, practical ridge-regression savings and the exact seen/unseen zero-shot probing protocol; it points to video demonstrations without replacing the protocol.

Official ZIP PDF supplement §§A–D, pp. 1–2, Table 6 · Main paper · Additional source 2

Clarity and writing conventions

The further-analysis section links each cue ablation to downstream tasks and shows an out-of-domain aerial-shadow failure, limiting claims that dense features universally transfer.

Tables 4–5 and Figures 6–7, PDF p. 8 · Main paper · Additional source 2

What to reuse—and what not to infer

Transferable lesson: For a representation objective, show the invariant mapping in the diagram and evaluate complementary downstream probes rather than only the pretext reconstruction loss.

Boundary: Unlabelled video training still uses pretrained depth/flow encoders and a pretrained visual backbone; zero-shot segmentation here includes a probe trained on seen classes, not no training of any kind.

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. 1–2, 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. Official ZIP PDF supplement §§A–D, pp. 1–2, Table 6. 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 supplementary ZIP PDF member: 35602_supp/camera-ready-supp.pdf; retrieved 2026-09-26 · 2 pages