|
|
|
| Semi-Supervised Semantic Segmentation Based on Generative Adversarial Networks |
| ZHU Feng, LIU Qipeng
|
| Institute of Complexity Science, Qingdao University, Qingdao 266071, China |
|
|
|
|
Abstract In this paper, we use generative adversarial network (GAN) to improve semantic segmentation of images. The model is composed of a semantic segmentation network and a discriminant network, where the segmentation network responses for generating semantic segmentation result while the discriminant network responses for detecting the difference between the generated result and the labels on the global structure level and improving the segmentation effect. In order to extract context information, we adopt the spatial pyramid pooling module in the segmentation network, which could perform pooling operation on multiple levels of sub-regions. Meanwhile, in order to solve the problem of a large number of manual annotations needed in the semantic segmentation data set, we use the discriminant network to generate pseudo labels and realize semi-supervision in the training of the segmentation network. The model has been tested using PASCAL VOC2012 dataset, and the results show that supervised and semi-supervised approaches proposed in this paper are superior to the existing methods.
|
|
Received: 21 July 2020
Published: 28 December 2020
|
|
|
|
|
|
| No related articles found! |
|
|
|
|
