scholarly journals Semantic segmentation of panoramic images using a synthetic dataset

Author(s):  
Yuanyou Xu ◽  
Kaiwei Wang ◽  
Kailun Yang ◽  
Dongming Sun ◽  
Jia Fu
Author(s):  
Y. Ao ◽  
J. Wang ◽  
M. Zhou ◽  
R. C. Lindenbergh ◽  
M. Y. Yang

<p><strong>Abstract.</strong> Panoramic images are widely used in many scenes, especially in virtual reality and street view capture. However, they are new for street furniture identification which is usually based on mobile laser scanning point cloud data or conventional 2D images. This study proposes to perform semantic segmentation on panoramic images and transformed images to separate light poles and traffic signs from background implemented by pre-trained Fully Convolutional Networks (FCN). FCN is the most important model for deep learning applied on semantic segmentation for its end to end training process and pixel-wise prediction. In this study, we use FCN-8s model that pre-trained on cityscape dataset and finetune it by our own data. The results show that in both pre-trained model and fine-tuning, transformed images have better prediction results than panoramic images.</p>


2020 ◽  
Vol 12 (3) ◽  
pp. 386 ◽  
Author(s):  
Rafał Wróżyński ◽  
Krzysztof Pyszny ◽  
Mariusz Sojka

The study presents a new method for quantitative landscape assessment. The method uses LiDAR data and combines the potential of GIS (ArcGIS) and 3D graphics software (Blender). The developed method allows one to create Classified Digital Surface Models (CDSM), which are then used to create 360° panoramic images from the point of view of the observer. In order to quantify the landscape, 360° panoramic images were transformed to the Interrupted Sinusoidal Projection using G.Projector software. A quantitative landscape assessment is carried out automatically with the following landscape classes: ground, low, medium, and high vegetation, buildings, water, and sky according to the LiDAR 1.2 standard. The results of the analysis are presented quantitatively—the percentage distribution of landscape classes in the 360° field of view. In order to fully describe the landscape around the observer, graphs of little planets have been proposed to interpret the obtained results. The usefulness of the developed methodology, together with examples of its application and the way of presenting the results, is described. The proposed Quantitative Landscape Assessment method (QLA360) allows quantitative landscape assessment to be performed in the 360° field of view without the need to carry out field surveys. The QLA360 uses LiDAR American Society of Photogrammetry and Remote Sensing (ASPRS) classification standards, which allows one to avoid differences resulting from the use of different algorithms for classifying images in semantic segmentation. The most important advantages of the method are as follows: observer-independent, 360° field of view which simulates human perspective, automatic operation, scalability, and easy presentation and interpretation of results.


2021 ◽  
Author(s):  
Anderson Brilhador ◽  
Matheus Gutoski ◽  
André Eugênio Lazzaretti ◽  
Heitor Silvério Lopes

Typical semantic segmentation methods do not recognize unknown pixels during the test or deployment stage. This capability is critical for open-world environment applications where unseen objects appear all the time. Recently, to solve those limitations, Open Set Semantic Segmentation (OSSS) was introduced. This task aims to produce known and unknown pixels semantic segments. However, due to its recent introduction, few works are found in the literature, and consequently, few datasets are publicly available. This work carried out a comparative study between the existing OSSS methods on a new synthetic dataset of images and the well-known PASCAL VOC 2012 dataset. The compared methods include SoftMax-T, OpenMax-based, and OpenIPCS. The results are encouraging and show some of the advantages and main limitations of each technique. However, in general, they demonstrate that the problem of OSSS remains open and demands further research aiming at real applications, such as autonomous driving and robotics.


Sensors ◽  
2019 ◽  
Vol 19 (11) ◽  
pp. 2577 ◽  
Author(s):  
Suvash Sharma ◽  
John E. Ball ◽  
Bo Tang ◽  
Daniel W. Carruth ◽  
Matthew Doude ◽  
...  

Since the state-of-the-art deep learning algorithms demand a large training dataset, which is often unavailable in some domains, the transfer of knowledge from one domain to another has been a trending technique in the computer vision field. However, this method may not be a straight-forward task considering several issues such as original network size or large differences between the source and target domain. In this paper, we perform transfer learning for semantic segmentation of off-road driving environments using a pre-trained segmentation network called DeconvNet. We explore and verify two important aspects regarding transfer learning. First, since the original network size was very large and did not perform well for our application, we proposed a smaller network, which we call the light-weight network. This light-weight network is half the size to the original DeconvNet architecture. We transferred the knowledge from the pre-trained DeconvNet to our light-weight network and fine-tuned it. Second, we used synthetic datasets as the intermediate domain before training with the real-world off-road driving data. Fine-tuning the model trained with the synthetic dataset that simulates the off-road driving environment provides more accurate results for the segmentation of real-world off-road driving environments than transfer learning without using a synthetic dataset does, as long as the synthetic dataset is generated considering real-world variations. We also explore the issue whereby the use of a too simple and/or too random synthetic dataset results in negative transfer. We consider the Freiburg Forest dataset as a real-world off-road driving dataset.


Author(s):  
R. L. Garcia ◽  
P. N. Happ ◽  
R. Q. Feitosa

Abstract. This paper reports the results of a study that aims to develop semi-automatic methods for assessing the degree of corrosion in industrial plant. We evaluated two fully convolutional networks (U-Net and DeepLab v3 +) to segment corroded areas in panoramic images of offshore platforms. The experimental analysis was based on two datasets built for this study. The datasets comprise 9,112 2D images and 3,732 panoramic images. Both FCNs trained on 2D images were tested on 2D images and cubic projections of panoramic images. In addition to pointing out encouraging results, the experiments indicated that most prediction errors concentrated in corrosion defects with a small pixel area.


2018 ◽  
Vol 11 (6) ◽  
pp. 304
Author(s):  
Javier Pinzon-Arenas ◽  
Robinson Jimenez-Moreno ◽  
Ruben Hernandez-Beleno

Sign in / Sign up

Export Citation Format

Share Document