ACDNet with ASPP for Camouflaged Object Detection

Abstract

This paper proposes a camouflage identification network ACDNet, which can perform end-to-end search and identification simply and efficiently. Use ASPP (Atrous Spatial Pyramid Pooling) to enhance the image features of camouflaged object, and then connect the Parallel Decoder (PD) module to get the global map, and then use the Group-Reversal Block (GRB) module to connect sequentially, refine the details, and finally perform deep supervision learning got result. Compared with the current 13 methods, ACDNet has a significant performance improvement in accuracy and similarity, and it is one of the potential solutions for camouflage identification.