Object Detection of Road Assets Using Transformer-Based YOLOX with Feature Pyramid Decoder on Thai Highway Panorama

Abstract

Due to the various sizes of each object, such as kilometer stones, detection is still a challenge, and it directly impacts the accuracy of these object counts. Transformers have demonstrated impressive results in various natural language processing (NLP) and image processing tasks due to long-range modeling dependencies. This paper aims to propose an exceeding you only look once (YOLO) series with two contributions: (i) We propose to employ a pre-training objective to gain the original visual tokens based on the image patches on road asset images. By utilizing pre-training Vision Transformer (ViT) as a backbone, we immediately fine-tune the model weights on downstream tasks by joining task layers upon the pre-trained encoder. (ii) We apply Feature Pyramid Network (FPN) decoder designs to our deep learning network to learn the importance of different input features instead of simply summing up or concatenating, which may cause feature mismatch and performance degradation. Conclusively, our proposed method (Transformer-Based YOLOX with FPN) learns very general representations of objects. It significantly outperforms other state-of-the-art (SOTA) detectors, including YOLOv5S, YOLOv5M, and YOLOv5L. We boosted it to 61.5% AP on the Thailand highway corpus, surpassing the current best practice (YOLOv5L) by 2.56% AP for the test-dev data set.

YOLOv4‐dense: A smaller and faster YOLOv4 for real‐time edge‐device based object detection in traffic scene

“…RFBnet is constructed by SSD with VGG as the backbone network. It mainly adds dilated convolution on the basis of inception, thus effectively increasing the receptive field. Faster‐RCNN: it removes the selective search to generate the region proposal instead of using RPN (region proposal network) to generate the region candidate box. YOLO series: we re‐implement the models including YOLOv3 [27], YOLOv4 [6],Scaled‐YOLOv4‐large‐p5 [27], PP‐YOLO‐r18vd [24], YOLOX‐s [25], YOLOv5s [26], YOLOX‐tiny [25], YOLOv3‐tiny [27], YOLOv4‐tiny [27] and PP‐YOLO‐tiny [24] using the same data splits and experimental setting.…”

“…Recently, edge-device based objects detection has been raised more attention form the community boosted by the huge demand from the hundreds of millions edge devices, such as, vehicles, smart phones etc. The typical edge-device based object detectors are PP-YOLO [24], YOLOX [25], YOLOv5s [26], YOLOv4-tiny [27]. PP-YOLO replaced part of the convolution layer with deformable convolution, deleted the dropout of the backbone part then changed it to the detection head part And use a matrix NMS that runs faster than traditional NMS.…”

Enhanced pothole detection system using YOLOX algorithm

“…It achieved 50.0% AP on COCO dataset at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. The system presented by Teerapong et al [42] uses YOLOX algorithm to detect road assets on the highways. To avoid using multiple cameras to detect objects from all four sides of the vehicle, the authors developed a solution that uses a 360-degree spherical camera to detect objects and other approaching vehicles.…”

Rapid Detection of Wheat Ears in Orthophotos From Unmanned Aerial Vehicles in Fields Based on YOLOX

“…Currently it is not used much in articles. Panboonyuen et al (2021) utilizing pre-training Vision Transformer (ViT) as a backbone, apply Feature Pyramid Network (FPN) decoder detection of Road Assets, It significantly outperforms other state-of-the-art (SOTA) detectors. Zhang et al (2021) used the YOLOX algorithm to detect vehicle targets in UAV images, and through a self-made dataset, the detection results surpassed traditional algorithms.…”