Year: 2025
Type: article
Abstract: Semantic segmentation has been one of the must research topics in the field of computer vision in recent years. This study was conducted using U-Net architecture in the context of self-driving cars on a cityscape dataset. The U-Net model uses a pre-trained RestNet101 for its encoder for feature extraction, and having skip connections between the encoder and the decoder with RELU activation. The skip connection helps to retain spatial information after downsampling, this enable the model to combined deep layers and fine-grained features. The model achieved 88% accuracy, 75.32% pixel accuracy, 72% precision, 80% recall and 75.79% F1 score metric for 75 epochs for over 2 hours and 30 minutes training time. The model has shown good performance on not only other state-of-the-art medical imaging papers but also on the cityscape dataset various driving scenarios in the context of autonomous driving. Therefore, we concluded that the U-Net model with RestNet101 encoders performed well in a typical complex scene environment on the cityscape dataset.
Authors: Jing Huang, Ebou A. Sowe
Institution: Wuhan University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4408955202