Performance Evaluation of Deep Learning Networks for Semantic Segmentation of Traffic Stereo-Pair Images

Abstract

Semantic image segmentation is one the most demanding task, especially for analysis of traffic conditions for self-driving cars. Here the results of application of several deep learning architectures (PSPNet and ICNet) for semantic image segmentation of traffic stereo-pair images are presented. The images from Cityscapes dataset and custom urban images were analyzed as to the segmentation accuracy and image inference time. For the models pre-trained on Cityscapes dataset, the inference time was equal in the limits of standard deviation, but the segmentation accuracy was different for various cities and stereo channels even. The distributions of accuracy (mean intersection over union -mIoU) values for each city and channel are asymmetric, long-tailed, and have many extreme outliers, especially for PSPNet network in comparison to ICNet network. Some statistical properties of these distributions (skewness, kurtosis) allow us to distinguish these two networks and open the question about relations between architecture of deep learning networks and statistical distribution of the predicted results (mIoU here). The results obtained demonstrated the different sensitivity of these networks to: (1) the local street view peculiarities in different cities that should be taken into account during the targeted fine tuning the models before their practical applications, (2) the right and left data channels in stereo-pairs. For both networks, the difference in the predicted results (mIoU here) for the right and left data channels in stereo-pairs is out of the limits of statistical error in relation to mIoU values. It means that the traffic stereo pairs can be effectively used not only for depth calculations (as it is usually used), but also as an additional data channel that can provide much more information about scene objects than simple duplication of the same street view images. ACM Reference format:NA https://doi.org/NA

Lidar Annotation Is All You Need

“…As a model, we use PSPNet semantic segmentation model [7], which is well-studied by the computer vision community. PSPNet and its variations show good results for segmentation tasks in different domains as well as for road segmentation [28] [29]. Moreover, it has benchmarks on various datasets, which are useful for research purposes.…”

Impact of Ground Truth Annotation Quality on Performance of Semantic Image Segmentation of Traffic Conditions

“…In addition to this, there is the comparative analysis of ICNet and PSPNet performance with regard to several subsets of Cityscapes dataset including stereo-pair images taken by left and right cameras for different cities [14]. It was found that the distributions of the mIoU values for each city and channel are asymmetric, longtailed, and have many extreme outliers, especially for PSPNet network in comparison to ICNet network.…”

“…The whole workflow was carried out according to the scenarios 1 and 2 described in the Table 1, namely, for scenario 1 the fine GT images were used for training and prediction, and for scenario 2 the fine GT images were used for training and the coarse GT images were used for prediction. It should be noted that scenario 1 was already implemented in our previous work [14] and scenarios 3 4 are under work right now and will be published elsewhere [18]. In this, work the following main classes, which could be used in the autonomous driving tasks, were selected: a road, a car, a person/pedestrian, traffic lights and signs classes.…”