Learning Discriminative Representations From Cross-Scale Features for Camouflaged Object Detection

The key that hinders the performance improvement of current camouflaged object detection (COD) models is the lack of discriminability of features at fine granularity. We solve this problem from two complementary perspectives. Firstly, complex scenes result in the discriminative feature representations of camouflaged objects being present at different scales and semantic abstraction levels. Therefore, a mechanism is needed to increase the diversity of features to integrate more information potentially beneficial for COD. Second, appearance similarity between objects and environments will inevitably lead to similarity in features. Enhancing feature diversity alone is not enough to solve the above problems. Therefore, it is necessary to give the model semantic perception capabilities to expand the subtle discrepancies between objects and environments in feature embedding. Inspired by the first point, we propose a cross-scale interaction module (CSIM) that utilizes cross-attention between different scales to enhance the diversity of feature representations. Regarding the second point, the semantic guided feature learning (SGFL) is proposed to promote the model to expand feature discrepancies through explicit supervision. Experiments on four popular COD datasets show that our method outperforms recent SOTA methods. In addition, polyp segmentation experiments show that it is also effective for other COD-like tasks.

I. Introduction

Camouflage is a common phenomenon in nature. Animals often blend into their surroundings to confuse prey or hide from predators [1]. For example, to avoid predators, the skin of the leaf-tailed gecko is covered with bumps that mimic the texture of tree bark. Alligator snapping turtles, which have a dark brown or black shell with a series of sharp protrusions, often hide in the mud and wait for prey to approach. In human society, COD also has broad application prospects, such as medical image analysis, agricultural pest detection, architectural design, species conservation [2], [3]. The purpose of COD is to find these objects hidden in the surroundings [4]. However, compared with traditional object detection or segmentation tasks, COD has more challenges, which can be observed in Fig. 1. Firstly, the contrast between the object and the background may not be very strong. Even if the object occupies a large proportion in the image, it is still difficult to detect the object completely, as shown in row 1 of Fig. 1. This situation often results in failed detection of local parts of the object. Secondly, the complex background environment and the small size of the target objects further complicate the task. The imbalance of pixels between small targets and the environment greatly affects the detection effect, resulting in unclear detection of object details, as shown in row 2 of Fig. 1. Thirdly, the detection of multiple camouflage objects often leads to missed detections, as shown in row 3 of Fig. 1. Finally, occlusion between objects is also one of the main factors leading to incomplete detection of camouflaged objects, as illustrated in row 4 of Fig. 1. Fig. 1.

Visual examples of camouflaged object detection results from different methods in challenging scenes. Row 1: camouflaged detection of large-sized objects that are highly similar to the background. Row 2: camouflaged detection of small-sized objects hidden in the environment. Row 3: Multi-target camouflaged object detection. Row 4: Occluded camouflaged object detection. (Best viewed digitally.)