CARL-ODD: A Vision Benchmark Dataset of Asia for On-Road Vehicle Detection and Recognition - Figure 8

Two-stage object detectors are beneficial when high accuracy is a crucial factor compared to the inference speed, for example, in bioinformatics applications and text recognition. Therefore, to demonstrate the diversity  of our proposed dataset, this article employed a widely recognised two-stage object detector called Faster Region-Based Convolutional Neural Network (FasterR-CNN). This network architecture consists of three  neural networks: the first is the Feature Extraction Network, followed by Region Proposal Network, and Detection Network. The proposed architecture is shown in Figure 8.

In the benchmark data, the deployed single-stage object detector achieved a mean average precision of 84.40% (Figure 8), demonstrating satisfactory performance for use in machine learning applications where real-time detection is critical.