OSFormer: One-Stage Camouflaged Instance Segmentation with Transformers

We present OSFormer, the first one-stage transformer framework for camouflaged instance segmentation (CIS). OSFormer is based on two key designs. First, we design a location-sensing (LST) to obtain location label and instance-aware parameters by introducing location-guided queries blend-convolution feed-forward network. Second, develop coarse-to-fine fusion (CFF) merge diverse context information from LST encoder CNN backbone. Coupling these components enables efficiently blend local features long-range dependencies predicting instances. Compared with two-stage frameworks, our reaches 41% AP achieves good convergence efficiency without requiring enormous training data, i.e., only 3,040 samples under 60 epochs. Code link: https://github.com/PJLallen/OSFormer .