RetCompletion:High-Speed Inference Image Completion with Retentive Network
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917856337199104 |
|---|---|
| author | Cang, Yueyang Hu, Pingge Zhang, Xiaoteng Wang, Xingtong Liu, Yuhang Shi, Li |
| author_facet | Cang, Yueyang Hu, Pingge Zhang, Xiaoteng Wang, Xingtong Liu, Yuhang Shi, Li |
| contents | Time cost is a major challenge in achieving high-quality pluralistic image completion. Recently, the Retentive Network (RetNet) in natural language processing offers a novel approach to this problem with its low-cost inference capabilities. Inspired by this, we apply RetNet to the pluralistic image completion task in computer vision. We present RetCompletion, a two-stage framework. In the first stage, we introduce Bi-RetNet, a bidirectional sequence information fusion model that integrates contextual information from images. During inference, we employ a unidirectional pixel-wise update strategy to restore consistent image structures, achieving both high reconstruction quality and fast inference speed. In the second stage, we use a CNN for low-resolution upsampling to enhance texture details. Experiments on ImageNet and CelebA-HQ demonstrate that our inference speed is 10$\times$ faster than ICT and 15$\times$ faster than RePaint. The proposed RetCompletion significantly improves inference speed and delivers strong performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_04056 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | RetCompletion:High-Speed Inference Image Completion with Retentive Network Cang, Yueyang Hu, Pingge Zhang, Xiaoteng Wang, Xingtong Liu, Yuhang Shi, Li Computer Vision and Pattern Recognition Time cost is a major challenge in achieving high-quality pluralistic image completion. Recently, the Retentive Network (RetNet) in natural language processing offers a novel approach to this problem with its low-cost inference capabilities. Inspired by this, we apply RetNet to the pluralistic image completion task in computer vision. We present RetCompletion, a two-stage framework. In the first stage, we introduce Bi-RetNet, a bidirectional sequence information fusion model that integrates contextual information from images. During inference, we employ a unidirectional pixel-wise update strategy to restore consistent image structures, achieving both high reconstruction quality and fast inference speed. In the second stage, we use a CNN for low-resolution upsampling to enhance texture details. Experiments on ImageNet and CelebA-HQ demonstrate that our inference speed is 10$\times$ faster than ICT and 15$\times$ faster than RePaint. The proposed RetCompletion significantly improves inference speed and delivers strong performance. |
| title | RetCompletion:High-Speed Inference Image Completion with Retentive Network |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2410.04056 |