GenCAMO: Scene-Graph Contextual Decoupling for Environment-aware and Mask-free Camouflage Image-Dense Annotation Generation

TL;DR

This work tackles the scarcity of dense camouflage data by introducing GenCAMO, a mask-free, environment-aware generation framework guided by scene-graphs. It builds GenCAMO-DB, a large multi-modal dataset with depth maps, scene graphs, and prompts, enabling controllable camouflage synthesis. The GenCAMO architecture combines Depth–Layout Coherence (DLCG-ControlNet) and Attribute-aware Mask Attention (AMA) within a unified diffusion framework, achieving geometry-consistent, context-aware generation and dense annotations (image, depth, mask). Extensive experiments on CIG and S2RCDP tasks show that synthetic data from GenCAMO improves both generation quality (FID/KID) and downstream dense-prediction performance, reducing the synthetic–real gap and enhancing robustness in mask-scarce environments.

Abstract

Conceal dense prediction (CDP), especially RGB-D camouflage object detection and open-vocabulary camouflage object segmentation, plays a crucial role in advancing the understanding and reasoning of complex camouflage scenes. However, high-quality and large-scale camouflage datasets with dense annotation remain scarce due to expensive data collection and labeling costs. To address this challenge, we explore leveraging generative models to synthesize realistic camouflage image-dense data for training CDP models with fine-grained representations, prior knowledge, and auxiliary reasoning. Concretely, our contributions are threefold: (i) we introduce GenCAMO-DB, a large-scale camouflage dataset with multi-modal annotations, including depth maps, scene graphs, attribute descriptions, and text prompts; (ii) we present GenCAMO, an environment-aware and mask-free generative framework that produces high-fidelity camouflage image-dense annotations; (iii) extensive experiments across multiple modalities demonstrate that GenCAMO significantly improves dense prediction performance on complex camouflage scenes by providing high-quality synthetic data. The code and datasets will be released after paper acceptance.

Figures (13)

• Figure 1: Motivation, Method, and Application of this work. Inspired by the environment-aware camouflage abilityteyssier2015photonic of chameleons, GenCAMO is a mask-free generative framework that takes image–text–depth conditions as input and produces realistic and context-adaptive camouflage images together with their depth and mask annotations. These outputs are guided by a scene-graph decoupling mechanism that separates object attributes, relations, and environmental cues to achieve controllable generation.
• Figure 2: Illustration of the semantic concepts distribution for the concealed, salient and general categories in our GenCAMO-DB.
• Figure 3: Overview of our dataset construction pipeline. Depth maps, scene graphs, and captions are automatically generated for 34K images, followed by human verification and refinement.
• Figure 4: Overview of the proposed method framework. GenCAMO integrates visual, textual, and scene-graph cues through semantic–layout decoupling, depth–layout coherence guidance, and attribute-aware mask attention to generate context-adaptive camouflage images with corresponding depth and mask annotations.
• Figure 5: Multi-modal controllable camouflage image synthesis. Comparison of LAKE-RED, MIP-Adapter, and GenCAMO under Text + Image, Depth + Image, and Text + Depth + Image with the scene graph.