Multi-Loss-based Semi-Supervised Image Semantic Segmentation And Its Extensions Based On Deep Learning Techniques

Image semantic segmentation is popular in computer vision,since the high-level semantic understanding of images can be effectively realized.Based on whether and to what extent the training data are labeled,most image semantic segmentation methods can be categorized into fully-supervised learning-based methods,weakly-supervised learning-based methods,and semi–supervised learning-based methods.Among them,semi-supervised image semantic segmentation receives increasing popularity recently,because of its flexibility and convenience in requiring partial training data to be labeled.Although semi-supervised image semantic segmentation is promising,the existing methods are not effective in dealing with images with special conditions such as poor lighting conditions,segmentation of small size objects and multiple objects with the same semantics.To deal with the above dilemma,a novel multi-loss-based deep adversarial network is proposed in this paper.Technically,the more robust WGAN-GP model is utilized as the backbone of the novel network,instead of the conventional GAN model.Moreover,multiple losses including the cross-entropy loss,the edge detection loss,the adversarial loss,and the semi-supervised loss,are incorporated during the novel network's training.Experimental analysis based on challenging cases selected from the famous Pascal Voc 2012 data set and Cityscapes data set show that the method improves the detail handling of the segmentation model.Furthermore,this paper continues to design comparative experiments to explore the impact of the proportion of labeled data in the data set on the segmentation performance of the model in semi supervised learning.Because the generator proposed in this paper can be any segmentation model,the existing FCN,deep lab and other fully supervised training segmentation models are replaced by the original generator,which makes full supervised learning more efficient In this way,we can reduce the heavy work of pixel level labeling of data sets.After using Pascal Voc2012 and cityscapes data sets for comparative experiments,it is found that with the increase of labeled data,the model will train better,but the full supervision training may not get the best semantic segmentation results.Through the research of Multi-loss-based semi-supervised image semantic segmentation and its extensions based on deep learning techniques,it can be found that semi-supervised learning contains great advantages.