In this project, I conducted an experimental analysis of YOLOv3's capability for real-time object detection, a critical component in autonomous driving systems. After an extensive review and comparison of leading real-time object detection algorithms, including SSD and previous YOLO versions, YOLOv3 emerged as the superior choice in terms of speed and accuracy. However, challenges remain in detecting small objects and addressing class imbalances, as evidenced by YOLOv3's performance on the BDD100k dataset.
While YOLOv3 showed a promising mAP of 15.44 overall, it particularly excelled with a 32.96 AP in detecting cars, albeit with difficulties in recognizing distant vehicles and those obscured by headlights. Pedestrian detection also proved problematic, especially at farther ranges. To improve upon these issues, data augmentation strategies were proposed, with further suggestions to utilize more balanced datasets like KITTI for training.
[Read Full Post]
For this project, I propose an innovative approach to optimize anchor boxes for the state-of-the-art object detection algorithm Single Shot Detector. Traditionally, anchor boxes are predefined in a simplistic, hand-picked manner. For instance, Wei et al. selected anchors with ratios of 1:1, 1:2, 2:1, 1:3, and 3:1, and their implementation of Single Shot Detector 300x300 achieved a 74.3% mAP on the VOC2007 testing set.
To enhance accuracy and streamline the model, my proposal involves using k-means clustering on the ground truth boxes of the VOC2007+VOC2012 training set. This method is aimed at tailoring the anchor boxes more precisely to the data distribution, thereby identifying the most effective number of anchor boxes and their aspect ratios. Through this approach, it was discovered that the number of anchors per grid cell could be reduced from 6 to 4 while still achieving an increase in mAP by 0.7%. This refinement not only improves performance but also makes the model more efficient and data-adaptive.
[Read Full Post]
