REF-Net: Robust, Efficient, and Fast Network for Semantic Segmentation Applications Using Devices With Limited Computational Resources

Abstract

Considering importance of the autonomous driving applications for mobile devices, it is imperative to develop both fast and accurate semantic segmentation models. Thanks to emergence of Deep Learning (DL) techniques, the segmentation models enhanced their accuracy. However, this improved performance of currently popular DL models for self-driving car applications come at the cost of time and computational efficiency. Moreover, networks with efficient model architecture experience lack of accuracy. Therefore, in this study, we propose robust, efficient, and fast network (REF-Net) that combines carefully formulated encoding and decoding paths. Specifically, the contraction path uses mixture of dilated and asymmetric convolution layers with skip connections and bottleneck layers, while the decoding path benefits from nearest neighbor interpolation method that demands no trainable parameters to restore original image size. This model architecture considerably reduces the number of trainable parameters, required memory space, training, and inference time. In fact, the proposed model required nearly 90 times fewer trainable parameters and approximately 4 times less memory space that allowed 3-fold faster training runtime and 2-fold inference speedup in the conducted experiments using Cambridge-driving Labeled Video Database (CamVid) and Cityscapes datasets. Moreover, despite its notable efficiency in terms of memory and time, the REF-Net attained superior results in several segmentation evaluation metrics that showed roughly 2%, 4%, and 3% increase in pixel accuracy, Dice coefficient, and Jaccard Index, respectively. INDEX TERMS Autonomous driving, deep convolutional neural networks, nearest neighbor interpolation, semantic segmentation.

An integrated mediapipe-optimized GRU model for Indian sign language recognition

“…After stage 1, the extracted features, that is, the landmark points ( ) per frame are flattened, concatenated and stored in a file to check and remove any null entries from the data. Data cleaning is important since it prevents failed detection of features 56 – 58 , which occurs when a blurred image is sent to the detector and leads to a null entry into the dataset. Thus, when training occurs with this noisy data, the prediction accuracy is reduced and bias may occur.…”

Consecutive multiscale feature learning-based image classification model

“…As digital image data are significantly large in volume and usually complex, sophisticated digital image analysis techniques, such as machine learning (ML) and deep learning (DL), have been used to efficiently handle them 2 . Several tasks have been performed to deal with digital images, such as image classification, semantic segmentation 3 – 5 , object detection 6 , 7 , and instance segmentation 8 , 9 . Image classification is crucial part of digital image analysis and a basic component of the other computer vision tasks because image classification models are used as a backbone for the abovementioned more advanced computer vision tasks 10 , 11 .…”

Learning and Adaptation From Minimum Samples With Heterogeneous Quality: An Investigation of Image Segmentation Networks on Natural Dataset

“…There are CNN based semantic segmentation networks capable to reduce the utilization of computational resources while training and inference. The semantic segmentation networks named FU-net [27] and REF-net [28] specifically designed to deploy applications in hardware's with limited computational resources. Eventhough these networks are computationally cheap, these networks are trained on large data samples.…”