CNN Model & Tuning for Global Road Damage Detection

Abstract

This paper provides a report on our solution including model selection, tuning strategy and results obtained for Global Road Damage Detection Challenge. This Big Data Cup Challenge was held as a part of IEEE International Conference on Big Data 2020. We assess single and multi-stage network architectures for object detection and provide a benchmark using popular state-of-the-art open-source PyTorch frameworks like Detectron2 and Yolov5. Data preparation for provided Road Damage training dataset, captured using smartphone camera from Czech, India and Japan is discussed. We studied the effect of training on a per country basis with respect to a single generalizable model. We briefly describe the tuning strategy for the experiments conducted on two-stage Faster R-CNN with Deep Residual Network (Resnet) and Feature Pyramid Network (FPN) backbone. Additionally, we compare this to a one-stage Yolov5 model with Cross Stage Partial Network (CSPNet) backbone. We show a mean F1 score of 0.542 on Test2 and 0.536 on Test1 datasets using a multi-stage Faster R-CNN model, with Resnet-50 and Resnet-101 backbones respectively. This shows the generalizability of the Resnet-50 model when compared to its more complex counterparts. Experiments were conducted using Google Colab having K80 and a Linux PC with 1080Ti, NVIDIA consumer grade GPU. A PyTorch based Detectron2 code to preprocess, train, test and submit the Avg F1 score to is made available at https://github.com/vishwakarmarhl/rdd2020

Road damage detection algorithm for improved YOLOv5

“…Compared with the original YOLOv5s algorithm, the mAP is increased by 2.5%, and the F1 score is increased by 2.6%. An F1 score of 61.86% and an F1 score of 60.92 were achieved on the two test sets, and the comparison with the results submitted on the Global Road Damage Detection Challenge'2020 is shown in Table 7 10, 15,[23][24][25] .…”

Improved YOLOv5-Based Real-Time Road Pavement Damage Detection in Road Infrastructure Management

“…Singh et al used Mask RCNN with the RDD 2018 dataset to acquire an F1 score of 52.8% [47]. Numerous studies utilized RDD 2020, the successor to RDD 2018, and produced a higher F1 score than previous research [48,49]. Using the RDD 2020 dataset and other object detection algorithms, the researcher discovered that the YOLO-based approach outperformed most of the other studies [27].…”

Forest Roads Damage Detection Based on Objected Detection Deep Learning Algorithms

“…Often there are multiple targets present in the same image for the deterioration detection system in forest road to detect. YOLO is a very simple, single convolutional network that simultaneously predicts several boundary boxes and class possibilities for those boxes [22]. This integrated model has several advantages over traditional object detection methods, because it uses a regression program in this algorithm, so there is no need to use a complex pipeline, and the entire neural network is simply used to detect objects in images.…”

“…Late we will implement a memory map for increasing detection accuracy in video and image for different gridding size's effect by applying hybrid YOLO on acquaintance estimation task. Some image pre-process has been introduced, for instance, propose a palm-print-based identification system in [22], the pre-processing steps including image thresholding, border tracing and wavelet-based segmentation, the pre-processing method is proved to be effective and can be simulated in other scenarios as well. One of the key problems is how to express image data, which can usually be represented by features such as texture, colour, edge, shape [23].…”