@inproceedings{volkova-etal-2026-adapt,
title = "Adapt the Words, Freeze the Model: {LLM} Prompt Evolution for Domain-Shift-Robust Segmentation",
author = "Volkova, Olga and
Glazkov, Nikita and
Konstantin, Pchelin and
Bespalov, Iaroslav Radionovich and
Baraboshkin, Evgeny Evgenievich and
Orlov, Denis and
Dylov, Dmitry V.",
editor = "Atanasova, Pepa and
Haffari, Gholamreza and
Wang, Minghan",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-industry.214/",
pages = "3588--3601",
ISBN = "979-8-89176-460-6",
abstract = "In specialist industrial imagery the dominant cost is maintenance under domain shift: labeling and fine-tuning must be redone on every new field, and proprietary imagery often cannot leave the operator{'}s environment. For geological drill-core fracture segmentation we release CoreFrac, to our knowledge the first open pixel-level benchmark on drill-core photographs: 200 expert-annotated full-resolution images with an expert-agreement ceiling (soft-F1 0.86). We retarget a frozen concept segmentor (SAM 3) by evolving only its natural-language prompt {---} no gradient updates, just a small labeled probe. Under a source-grouped protocol the evolved prompt reaches 0.545 positives-only soft-F1, beating rank-8 LoRA at the same 96-patch cap (0.487; paired +0.058, 95{\%} CI [+0.020, +0.094]) and the best of eight expert prompts by 2.7{\texttimes}. Under a controlled domain shift, re-evolving on a small probe lifts a 0.622 {\textrightarrow} 0.409 drop back to 0.534. Its masks distill into a small on-premises U-Net (7.8M parameters, 3.8 ms/patch) that retains {\ensuremath{\approx}}92{\%} of the teacher{'}s positives-only soft-F1 (0.501) at {\ensuremath{\approx}}1700{\texttimes} lower latency than *unbatched* SAM 3 {---} a cheaper, leak-free deployment. Code and data are released."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="volkova-etal-2026-adapt">
<titleInfo>
<title>Adapt the Words, Freeze the Model: LLM Prompt Evolution for Domain-Shift-Robust Segmentation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Olga</namePart>
<namePart type="family">Volkova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikita</namePart>
<namePart type="family">Glazkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pchelin</namePart>
<namePart type="family">Konstantin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Iaroslav</namePart>
<namePart type="given">Radionovich</namePart>
<namePart type="family">Bespalov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Evgeny</namePart>
<namePart type="given">Evgenievich</namePart>
<namePart type="family">Baraboshkin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Denis</namePart>
<namePart type="family">Orlov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dmitry</namePart>
<namePart type="given">V</namePart>
<namePart type="family">Dylov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pepa</namePart>
<namePart type="family">Atanasova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gholamreza</namePart>
<namePart type="family">Haffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minghan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-460-6</identifier>
</relatedItem>
<abstract>In specialist industrial imagery the dominant cost is maintenance under domain shift: labeling and fine-tuning must be redone on every new field, and proprietary imagery often cannot leave the operator’s environment. For geological drill-core fracture segmentation we release CoreFrac, to our knowledge the first open pixel-level benchmark on drill-core photographs: 200 expert-annotated full-resolution images with an expert-agreement ceiling (soft-F1 0.86). We retarget a frozen concept segmentor (SAM 3) by evolving only its natural-language prompt — no gradient updates, just a small labeled probe. Under a source-grouped protocol the evolved prompt reaches 0.545 positives-only soft-F1, beating rank-8 LoRA at the same 96-patch cap (0.487; paired +0.058, 95% CI [+0.020, +0.094]) and the best of eight expert prompts by 2.7×. Under a controlled domain shift, re-evolving on a small probe lifts a 0.622 → 0.409 drop back to 0.534. Its masks distill into a small on-premises U-Net (7.8M parameters, 3.8 ms/patch) that retains \ensuremath\approx92% of the teacher’s positives-only soft-F1 (0.501) at \ensuremath\approx1700× lower latency than *unbatched* SAM 3 — a cheaper, leak-free deployment. Code and data are released.</abstract>
<identifier type="citekey">volkova-etal-2026-adapt</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-industry.214/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>3588</start>
<end>3601</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Adapt the Words, Freeze the Model: LLM Prompt Evolution for Domain-Shift-Robust Segmentation
%A Volkova, Olga
%A Glazkov, Nikita
%A Konstantin, Pchelin
%A Bespalov, Iaroslav Radionovich
%A Baraboshkin, Evgeny Evgenievich
%A Orlov, Denis
%A Dylov, Dmitry V.
%Y Atanasova, Pepa
%Y Haffari, Gholamreza
%Y Wang, Minghan
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-460-6
%F volkova-etal-2026-adapt
%X In specialist industrial imagery the dominant cost is maintenance under domain shift: labeling and fine-tuning must be redone on every new field, and proprietary imagery often cannot leave the operator’s environment. For geological drill-core fracture segmentation we release CoreFrac, to our knowledge the first open pixel-level benchmark on drill-core photographs: 200 expert-annotated full-resolution images with an expert-agreement ceiling (soft-F1 0.86). We retarget a frozen concept segmentor (SAM 3) by evolving only its natural-language prompt — no gradient updates, just a small labeled probe. Under a source-grouped protocol the evolved prompt reaches 0.545 positives-only soft-F1, beating rank-8 LoRA at the same 96-patch cap (0.487; paired +0.058, 95% CI [+0.020, +0.094]) and the best of eight expert prompts by 2.7×. Under a controlled domain shift, re-evolving on a small probe lifts a 0.622 → 0.409 drop back to 0.534. Its masks distill into a small on-premises U-Net (7.8M parameters, 3.8 ms/patch) that retains \ensuremath\approx92% of the teacher’s positives-only soft-F1 (0.501) at \ensuremath\approx1700× lower latency than *unbatched* SAM 3 — a cheaper, leak-free deployment. Code and data are released.
%U https://aclanthology.org/2026.emnlp-industry.214/
%P 3588-3601
Markdown (Informal)
[Adapt the Words, Freeze the Model: LLM Prompt Evolution for Domain-Shift-Robust Segmentation](https://aclanthology.org/2026.emnlp-industry.214/) (Volkova et al., EMNLP 2026)
ACL
- Olga Volkova, Nikita Glazkov, Pchelin Konstantin, Iaroslav Radionovich Bespalov, Evgeny Evgenievich Baraboshkin, Denis Orlov, and Dmitry V. Dylov. 2026. Adapt the Words, Freeze the Model: LLM Prompt Evolution for Domain-Shift-Robust Segmentation. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 3588–3601, Budapest, Hungary. Association for Computational Linguistics.