CamoTeacher: Dual-Rotation Consistency Learning for Semi-Supervised Camouflaged Object Detection

Xunfa Lai,Zhiyu Yang,Jie Hu,Shengchuan Zhang,Liujuan Cao,Guannan Jiang,Zhiyu Wang,Songan Zhang,Rongrong Ji

2024-08-15

Abstract:Existing camouflaged object detection~(COD) methods depend heavily on large-scale pixel-level annotations.However, acquiring such annotations is laborious due to the inherent camouflage characteristics of the objects.Semi-supervised learning offers a promising solution to this challenge.Yet, its application in COD is hindered by significant pseudo-label noise, both pixel-level and instance-level.We introduce CamoTeacher, a novel semi-supervised COD framework, utilizing Dual-Rotation Consistency Learning~(DRCL) to effectively address these noise issues.Specifically, DRCL minimizes pseudo-label noise by leveraging rotation views' consistency in pixel-level and instance-level.First, it employs Pixel-wise Consistency Learning~(PCL) to deal with pixel-level noise by reweighting the different parts within the pseudo-label.Second, Instance-wise Consistency Learning~(ICL) is used to adjust weights for pseudo-labels, which handles instance-level noise.Extensive experiments on four COD benchmark datasets demonstrate that the proposed CamoTeacher not only achieves state-of-the-art compared with semi-supervised learning methods, but also rivals established fully-supervised learning methods.Our code will be available soon.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in Camouflaged Object Detection (COD), existing methods are highly dependent on large - scale pixel - level annotated data, and obtaining these annotated data is very labor - intensive and costly. Due to the characteristic that camouflaged objects are highly integrated with the background, it is particularly difficult to obtain high - quality annotated data. Therefore, the paper proposes a semi - supervised learning framework to reduce the dependence on annotated data and improve the performance of camouflaged object detection.

Specifically, the paper proposes CamoTeacher, a novel semi - supervised camouflaged object detection framework, which effectively deals with the problem of pseudo - label noise by introducing Dual - Rotation Consistency Learning (DRCL). DRCL minimizes pseudo - label noise at the pixel - level and instance - level by using the consistency of rotated views. The specific mechanisms include:

1. **Pixel - wise Consistency Learning (PCL)**: Deals with pixel - level noise by re - weighting different parts within the pseudo - label.
2. **Instance - wise Consistency Learning (ICL)**: Deals with instance - level noise by adjusting the weight of the pseudo - label.

The combination of these techniques makes CamoTeacher not only reach the state - of - the - art level among semi - supervised learning methods, but also in some cases can even be comparable to the performance of fully - supervised learning methods. Experimental results show that on multiple COD benchmark datasets, CamoTeacher significantly outperforms existing semi - supervised and weakly - supervised methods with only a small amount of annotated data, and even approaches or exceeds the performance of fully - supervised methods.