29 April 2024 in IEEE Access
The vision transformer (ViT) with global self-attention exhibits quadratic computational complexity that depends on the image size. To address this issue, window-based self-attention ViT limits attention area to a specific window, thereby mitigating the computational complexity. However, it cannot effectively capture the relationships between windows. The Swin Transformer, a representative window-based self-attention ViT, introduces shifted-window multi-head self-attention (SW-MSA) to capture the cross-window information. However, SW-MSA groups tokens that are close to each other in the image into one window and thus cannot capture relationships between distant tokens. Therefore, this paper introduces a feature-space window attention transformer (FSwin Transformer) that includes distant but similar tokens in one window. The proposed FSwin Transformer clusters similar tokens based on the feature space and conducts self-attention within the cluster. Thus, this approach helps understand the global context of the image by compensating for interactions between long-distance tokens, which cannot be captured when windows are set based on the image space. In addition, we incorporate a feature-space refinement method with channel and spatial attention to emphasize key parts and suppress non-essential parts. The refined feature map improves the representation power of the model, resulting in improved classification performance. Consequently, in classification tasks for ImagaNet-1K, FSwin Transformer outperforms existing transformer-based backbones, including the Swin Transformer.
6 November 2024 in IEEE Access
Vehicle-in-the-loop simulation (VILS) is a vehicle-testing technique that integrates high-fidelity simulation environments with real-world vehicles. Among existing simulation approaches for evaluating autonomous driving systems (ADS), VILS is particularly noteworthy because it faithfully reflects the dynamic characteristics of real-world vehicles and ensures repeatable and reproducible testing in diverse virtual scenarios. While researchers strive to implement a VILS platform that closely approximates real-world vehicle-testing environments, the performance of vehicles in VILS testing may differ from that observed in real-world testing, depending on the platform’s reliability. Therefore, methods must be established to validate the reliability of VILS platforms. Herein, we present the essential components of a VILS platform for evaluating ADS (AD-VILS) and propose a metho dology to validate the reliability of the implemented AD-VILS platform. This methodology includes scenario definition, techniques for VILS testing and real-world vehicle testing, and procedures for evaluating consistency and correlation based on statistical and mathematical comparisons between the datasets from virtual and real-world tests. Moreover, we empirically derive reliability evaluation criteria through iterative testing. This methodology aims to enhance the precision and reliability of ADS evaluations conducted on AD-VILS platforms.