References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: iSegFormer: Interactive Segmentation via Transformers with Application to 3D Knee MR Images
(Submitted on 21 Dec 2021 (v1), revised 18 Mar 2022 (this version, v5), latest version 17 Jul 2022 (v6))
Abstract: We propose iSegFormer, a memory-efficient transformer that combines a Swin transformer with a lightweight multilayer perceptron (MLP) decoder. With the efficient Swin transformer blocks for hierarchical self-attention and the simple MLP decoder for aggregating both local and global attention, iSegFormer learns powerful representations while achieving high computational efficiencies. Specifically, we apply iSegFormer to interactive 3D medical image segmentation.
Submission history
From: Qin Liu [view email][v1] Tue, 21 Dec 2021 16:16:30 GMT (167kb,D)
[v2] Wed, 26 Jan 2022 01:17:22 GMT (167kb,D)
[v3] Mon, 31 Jan 2022 22:27:52 GMT (168kb,D)
[v4] Thu, 3 Mar 2022 04:20:51 GMT (399kb,D)
[v5] Fri, 18 Mar 2022 03:29:48 GMT (1696kb,D)
[v6] Sun, 17 Jul 2022 13:20:09 GMT (785kb,D)
Link back to: arXiv, form interface, contact.