GFR-SAM: Training-Free Referring Camouflaged Object Segmentation via Cross-Image Prompting

Jul 16, 2026

Referring Camouflaged Object Detection (Ref-COD) demands accurate segmentation of hidden targets using reference cues, yet supervised training is annotation-intensive and training-free point-prompting is vulnerable to prompt localization errors. GFR-SAM introduces a robust training-free three-stage pipeline that replaces fragile point matching with a Generate-Filter-Refine flow. It generates candidate masks via cross-image in-context exemplar guidance for SAM3, filters distractors through region-global contrastive ranking, then refines boundaries by geometric-semantic integration using box and text prompts. Evaluated on R2C7K, it improves weighted F-measure by 8.7% over prior training-free methods and rivals supervised state of the art.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

Introduction

Problem: Referring camouflaged object detection (Ref-COD)

Limitations of supervised and point-prompt training-free methods

Proposed approach: GFR-SAM and the Generate-Filter-Refine pipeline

GFR-SAM: Training-Free Referring Camouflaged Object Segmentation via Cross-Image Prompting Yilong Yang∗ [Yilong.Yang@xmu.edu.cn](Yilong.Yang@xmu.edu.cn) Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University Xiamen, China Shengchuan Zhang [zsc_2016@xmu.edu.cn](zsc_2016@xmu.edu.cn)[ ](zsc_2016@xmu.edu.cn)Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University Xiamen, China Jianxin Tian∗ [thetian666@stu.xmu.edu.cn](thetian666@stu.xmu.edu.cn)[ ](thetian666@stu.xmu.edu.cn)Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University Xiamen, China Liujuan Cao† [caoliujuan@xmu.edu.cn](caoliujuan@xmu.edu.cn)[ ](caoliujuan@xmu.edu.cn)Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University Xiamen, China arXiv :2607 . 11732v1 [ cs .CV] 13 Jul 2026 Abstract Referring Camouflaged Object Detection (Ref-COD) requires segmenting hidden targets guided by reference cues. While supervised methods are annotation-heavy and training-free approaches via sparse point-prompting are sensitive to localization errors, we propose GFR-SAM, a robust three-stage training-free framework. GFR-SAM shifts the paradigm from fragile point-matching to a"Generate-Filter-Refine" pipeline. First, we introduce In-Context Exemplar-guided Segmentation, empowering SAM3 with crossimage inference to generate candidate masks via holistic visual exemplars, bypassing its native intra-image constraints. Second, a Region-Global Contrastive Filtering module ranks candidates through DINOv3-based prototypical alignment, effectively suppressing background distractors. Finally, a Geometric-Semantic Refinement module synergizes bounding box and text prompts to recover fine-grained boundaries and enhance instance recall. Evaluated on the R2C7K benchmark, GFR-SAM outperforms existing training-free methods by 8.7% in weighted F-measure (􀀛􀁆􀁖) and competes with supervised state-of-the-art counterparts. Ultimately, this work underscores the potential of unlocking SAM3’s latent capability for cross-image In-Context prompting, establishing a robust, training-free paradigm that effectively bridges the gap between general-purpose foundation models and specialized, label-intensive perception tasks without the need for task-specific fine-tuning. ∗ Equal contribution.†Corresponding author. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission [and/or a fee. Request permissions from permissions@acm.org](and/or a fee. Request permissions from permissions@acm.org). MM’ 26, Rio de Janeiro, Brazil © 2026 Copyright held by the owner/author(s) . Publication rights licensed to ACM. ACM ISBN 978-x-xxxx-xxxx-x/YYYY/MM [https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX) CCS Concepts • Computing methodologies → Object detection; Image segmentation. Keywords Ref-COD, Training-Free, SAM, In-Context Learning ACM Reference Format: Yilong Yang, Jianxin Tian, Shengchuan Zhang, and Liujuan Cao. 2026. GFRSAM: Training-Free Referring Camouflaged Object Segmentation via CrossImage Prompting. In Proceedings of MM’ 26 . ACM, New York, NY, USA, 10 pages. [https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX) 1 Introduction Camouflaged object detection (COD) [5] segments objects embedded in natural surroundings by mim

FAQs

What problem does GFR-SAM address in referring camouflaged object segmentation?

GFR-SAM targets Ref-COD, where models must segment camouflaged targets guided by reference cues. It aims to overcome annotation-heavy supervision and the prompt sensitivity of existing training-free point-prompt methods.

How does GFR-SAM differ from prior training-free approaches?

GFR-SAM replaces fragile point-matching with a three-stage Generate-Filter-Refine pipeline. It decouples candidate generation, distractor suppression, and boundary recovery into distinct modules rather than relying on a single point-driven step.

What are the three stages of the proposed framework and what does each do?

First, In-Context Exemplar-guided Segmentation (Ice-Seg) uses cross-image exemplars to generate candidate masks with SAM3. Second, Region-Global Contrastive Filtering (RGCF) ranks candidates using DINOv3-based prototypical alignment to suppress background distractors. Third, Geometric-Semantic Refinement (GSR) combines bounding box and text prompts to recover fine-grained boundaries and improve instance recall.

Jul 16, 2026

Upload by Valentina