Writing study: What CLIP Knows but Cannot Say: Recovering Negation from Frozen Intermediate Features
← ECCV 2026 oral writing corpus
Chen-yi Lu, Yueh-Shao Chen, Somali Chaterji · Official ECCV 2026 oral record
Figure, table and section ledger
Editorial contribution type: Empirical analysis, diagnosis & auditing (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
- ≈ 15.15 page-equivalents
- Main figures / tables
- 7 / 6
- 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. Selected disputed heading locations also checked in rendered source pages.
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
Makes the failure concrete with affirmative versus negated captions, then separates intermediate compositional information from final visual alignment.
- 3 Method: divergence diagnosis, PeakPatch and joint training
The layer-wise diagnosis appears before the correction modules, and the experiment sequence returns to layer selection, ordinary retrieval and control perturbations.
- 4 Experiments and analysis; 5 Conclusion
The ablation reports negation MCQ, overall MCQ and retrieval separately; the joint model is not best on every individual column, preserving the trade-off.
Problem, claim and contribution
Makes the failure concrete with affirmative versus negated captions, then separates intermediate compositional information from final visual alignment.
Figure 1 and Section 1, pp.1–4 · Main paper
Argument and section order
The layer-wise diagnosis appears before the correction modules, and the experiment sequence returns to layer selection, ordinary retrieval and control perturbations.
Sections 3–4, pp.6–15 · Main paper
Figures, captions and visual explanation
A diagnostic curve labels build-up, peak and collapse along depth. A later heat map varies anchor and peak layers next to the component ablation table.
Figure 2, p.3; Figure 4 and Table 5, p.13 · Main paper
Tables, comparisons and evidence
The ablation reports negation MCQ, overall MCQ and retrieval separately; the joint model is not best on every individual column, preserving the trade-off.
Table 5, p.13; Table 6, p.14 · Main paper
Appendix and reproducibility
Main-text pointers assign further backbone comparisons, zero-shot classification and limitations to supplementary material; that separate source was not inspected.
Sections 4.1,4.4 and 5, pp.11,13–15 · Main paper
Clarity and writing conventions
The mechanism names and layer symbols recur from diagnosis to correction; the inference paragraph explicitly states there is no polarity-classification switch.
Sections 3.2–3.3, pp.6–10 · Main paper
What to reuse—and what not to infer
Transferable lesson: When a diagnostic motivates a repair, connect the selected layer or variable to an independent sweep and test preserved behavior as well as the targeted failure.
Boundary: Embedding plots and attention patterns are explanatory evidence, not proof of semantic reasoning in every setting. Dual-axis plots must label both scales clearly.
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
- 3, 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.