FocusDiffuser: Perceiving Local Disparities for Camouflaged Object Detection

TL;DR

This work presents FocusDiffuser, a conditional diffusion model for camouflaged object detection that treats detection as a noise-to-mask generation task conditioned on the input image. It introduces two specialized modules, Boundary-Driven LookUp (BDLU) and Cyclic Positioning (CP), to emphasize local detail and iteratively refine focus during denoising. Through an Image Conditional Encoder (ICE) and a diffusion backbone, FocusDiffuser achieves state-of-the-art results on COD benchmarks CAMO, COD10K, and NC4K, with strong qualitative performance in boundary delineation. The model demonstrates the potential of generative diffusion approaches to COD and offers insights for applying similar strategies to high-resolution segmentation and related vision tasks.

Abstract

Detecting objects seamlessly blended into their surroundings represents a complex task for both human cognitive capabilities and advanced artificial intelligence algorithms. Currently, the majority of methodologies for detecting camouflaged objects mainly focus on utilizing discriminative models with various unique designs. However, it has been observed that generative models, such as Stable Diffusion, possess stronger capabilities for understanding various objects in complex environments; Yet their potential for the cognition and detection of camouflaged objects has not been extensively explored. In this study, we present a novel denoising diffusion model, namely FocusDiffuser, to investigate how generative models can enhance the detection and interpretation of camouflaged objects. We believe that the secret to spotting camouflaged objects lies in catching the subtle nuances in details. Consequently, our FocusDiffuser innovatively integrates specialized enhancements, notably the Boundary-Driven LookUp (BDLU) module and Cyclic Positioning (CP) module, to elevate standard diffusion models, significantly boosting the detail-oriented analytical capabilities. Our experiments demonstrate that FocusDiffuser, from a generative perspective, effectively addresses the challenge of camouflaged object detection, surpassing leading models on benchmarks like CAMO, COD10K and NC4K.

Figures (7)

• Figure 1: Ideal illustration. General COD models (a) directly map images to camouflaged object maps. In contrast, FocusDiffuser (b) generates the detection results through a reverse denoising process, based on mined information from image.
• Figure 2: The framework of our FocusDiffuser. Instead of relying on the discriminative learning paradigm, our framework adopts a generative approach to guarantee reliability and generalizability. Please refer to Sec. \ref{['sec:FocusDiffuser']} for details.
• Figure 3: Details of Boundary-driven LookUp (BDLU) and Cyclic Positioning (CP). These modules are grafted onto the diffusion model and run in tandem with it. Zoom in for details.
• Figure 4: Visualization comparisons of noisy masks changing over time t under various scale factor $\boldsymbol{b}$ settings.
• Figure 5: Qualitative comparisons of predicted camouflaged maps with state-of-the-art models.