uploading code of DDT

DACM-Few-shot.pytorch

Code for Doubly deformable aggregation of covariance matrices for few-shot segmentation

Comparison of the proposed DACM (b) with the framework of existing state-of-the-art methods (a). There are three main differences. (1) We propose to utilizeGP-based kernel functions for similarity measurement. (2) We use covariance matricesinstead of cosine similarity as the cost volume. (3) A DDT module is designed foreffective cost volume aggregation.

Training

---

Test

---