${D}^{3}${ETOR}: ${D}$ebate-Enhanced Pseudo Labeling and Frequency-Aware Progressive ${D}$ebiasing for Weakly-Supervised Camouflaged Object ${D}$etection with Scribble Annotations

TL;DR

This work tackles weakly-supervised camouflaged object detection with scribble annotations by introducing a two-stage framework, D^3ETOR. The first stage uses adaptive entropy-driven point sampling and a multi-agent multimodal chain-of-thought debate to produce higher-quality SAM-generated pseudo masks. The second stage presents FADeNet, a frequency-aware debiasing network that fuses low- and high-frequency features via window-based cross-attention while dynamically mitigating scribble bias. Extensive experiments on CAMO, COD10K, and NC4K demonstrate state-of-the-art performance among scribble-supervised methods and competitive results with fully supervised COD models.

Abstract

Weakly-Supervised Camouflaged Object Detection (WSCOD) aims to locate and segment objects that are visually concealed within their surrounding scenes, relying solely on sparse supervision such as scribble annotations. Despite recent progress, existing WSCOD methods still lag far behind fully supervised ones due to two major limitations: (1) the pseudo masks generated by general-purpose segmentation models (e.g., SAM) and filtered via rules are often unreliable, as these models lack the task-specific semantic understanding required for effective pseudo labeling in COD; and (2) the neglect of inherent annotation bias in scribbles, which hinders the model from capturing the global structure of camouflaged objects. To overcome these challenges, we propose ${D}^{3}$ETOR, a two-stage WSCOD framework consisting of Debate-Enhanced Pseudo Labeling and Frequency-Aware Progressive Debiasing. In the first stage, we introduce an adaptive entropy-driven point sampling method and a multi-agent debate mechanism to enhance the capability of SAM for COD, improving the interpretability and precision of pseudo masks. In the second stage, we design FADeNet, which progressively fuses multi-level frequency-aware features to balance global semantic understanding with local detail modeling, while dynamically reweighting supervision strength across regions to alleviate scribble bias. By jointly exploiting the supervision signals from both the pseudo masks and scribble semantics, ${D}^{3}$ETOR significantly narrows the gap between weakly and fully supervised COD, achieving state-of-the-art performance on multiple benchmarks.

Figures (10)

• Figure 1: Visualization of task objectives across different detection paradigms. (a) Input raw image. (b) Generic object detection aims to locate and classify all identifiable objects (segmented with distinct colors). (c) Salient object detection focuses on identifying objects that naturally attract human attention (e.g., the stick highlighted in green). (d) Camouflaged object detection identifies visually inconspicuous objects that blend into the background, such as the small snake highlighted in red.
• Figure 2: As a general-purpose segmentation model, SAM struggles to meet the specific demands of the COD task. Consequently, its limited semantic discriminative capability often leads to suboptimal segmentation performance on camouflaged objects.
• Figure 3: The relative distances from ground truth mask and scribble pixels to their respective nearest object boundaries reveal a notable discrepancy in spatial distribution. It is evident that scribble annotations are biased toward central object regions, whereas mask labels exhibit a more uniform coverage across entire object areas.
• Figure 4: An overview of the proposed ${D}^{3}$ETOR framework for weakly-supervised camouflaged object detection, which consists of two stages: debate-enhanced pseudo labeling and frequency-aware progressive debiasing.
• Figure 5: Framework of our proposed ${D}^{3}$ETOR for weakly-supervised camouflaged object detection (WSCOD) with scribble annotations. In (a), candidate masks are first generated using visual-prompted SAM and then filtered through a multi-agent debate mechanism. Afterwards, images are decomposed into low-frequency and high-frequency components in (b), balancing global semantics and local details. These features are progressively fused via window-based cross-attention mechanism to refine segmentation results, while supervision strength across regions is dynamically adjusted across regions to mitigate harmful bias in scribble annotations.