期刊
2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021
卷 -, 期 -, 页码 8716-8725出版社
IEEE COMPUTER SOC
DOI: 10.1109/CVPR46437.2021.00861
关键词
-
资金
- Alibaba Innovative Research (AIR) Program
- Major Scientific Research Project of Zhejiang Lab [2019DB0ZX01]
The paper introduces a new method of encoding high-resolution binary grid masks using discrete cosine transform (DCT), named DCT-Mask. This method shows significant gains in various frameworks, backbones, datasets, and training schedules, with minimal impact on running speed. The success of DCT-Mask lies in its ability to achieve high-quality mask representation with low complexity.
Binary grid mask representation is broadly used in instance segmentation. A representative instantiation is Mask R-CNN which predicts masks on a 28 x 28 binary grid. Generally, a low-resolution grid is not sufficient to capture the details, while a high-resolution grid dramatically increases the training complexity. In this paper, we propose a new mask representation by applying the discrete cosine transform(DCT) to encode the high-resolution binary grid mask into a compact vector. Our method, termed DCT-Mask, could be easily integrated into most pixel-based instance segmentation methods. Without any bells and whistles, DCT-Mask yields significant gains on different frameworks, backbones, datasets, and training schedules. It does not require any pre-processing or pre-training, and almost no harm to the running speed. Especially, for higher-quality annotations and more complex backbones, our method has a greater improvement. Moreover, we analyze the performance of our method from the perspective of the quality of mask representation. The main reason why DCT-Mask works well is that it obtains a high-quality mask representation with low complexity.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据