Evaluation of deep learning algorithms for semantic segmentation of car parts

doi:10.1007/s40747-021-00397-8

Paper

Evaluation of deep learning algorithms for semantic segmentation of car parts

Published May 22, 2021 · Kitsuchart Pasupa, Phongsathorn Kittiworapanya, Napasin Hongngern

Complex & Intelligent Systems

UNKNOWN SJR score

37

Citations

5

Influential Citations

Full textSemantic Scholar

Abstract

Evaluation of car damages from an accident is one of the most important processes in the car insurance business. Currently, it still needs a manual examination of every basic part. It is expected that a smart device will be able to do this evaluation more efficiently in the future. In this study, we evaluated and compared five deep learning algorithms for semantic segmentation of car parts. The baseline reference algorithm was Mask R-CNN, and the other algorithms were HTC, CBNet, PANet, and GCNet. Runs of instance segmentation were conducted with those five algorithms. HTC with ResNet-50 was the best algorithm for instance segmentation on various kinds of cars such as sedans, trucks, and SUVs. It achieved a mean average precision at 55.2 on our original data set, that assigned different labels to the left and right sides and 59.1 when a single label was assigned to both sides. In addition, the models from every algorithm were tested for robustness, by running them on images of parts, in a real environment with various weather conditions, including snow, frost, fog and various lighting conditions. GCNet was the most robust; it achieved a mean performance under corruption, mPC = 35.2, and a relative degradation of performance on corrupted data, compared to clean data (rPC), of 64.4%, when left and right sides were assigned different labels, and mPC = 38.1 and rPC = $$69.6\%$$ when left- and right-side parts were considered the same part. The findings from this study may directly benefit developers of automated car damage evaluation system in their quest for the best design.

Study Snapshot

HTC with ResNet-50 is the best deep learning algorithm for semantic segmentation of car parts, with GCNet being the most robust and effective in real-world conditions.

PopulationOlder adults (50-71 years)

Sample size24

MethodsObservational

OutcomesBody Mass Index projections

ResultsSocial networks mitigate obesity in older groups.

Sign up to use Study Snapshot

Consensus is limited without an account. Create an account or sign in to get more searches and use the Study Snapshot.

Create an account

Paper

Evaluation of deep learning algorithms for semantic segmentation of car parts

Sign up to use Study Snapshot

References

Deep Learning-Based Car Damage Classification and Detection

Deep convolutional networks can accurately recognize vehicle damages, potentially accelerating insurance claims processing.

Neural Network for Low-Memory IoT Devices and MNIST Image Recognition Using Kernels Based on Logistic Map

LogNNet neural network effectively recognizes handwritten digits from MNIST-10 with 80.3-96.3% classification accuracy on low-memory IoT devices, offering a memory-efficient solution for AI applications.

SOLOv2: Dynamic and Fast Instance Segmentation

SOLOv2 is a simple, fast, and accurate instance segmentation framework with strong performance, outperforming state-of-the-art methods in both speed and accuracy.

BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation

BlendMask improves mask prediction in instance segmentation by combining top-down and bottom-up approaches, outperforming Mask R-CNN while being faster and simpler to integrate with detection frameworks.

RDSNet: A New Deep Architecture for Reciprocal Object Detection and Instance Segmentation

RDSNet effectively and efficiently combines object detection and instance segmentation in computer vision tasks, improving both accuracy and efficiency.

Citations

Real-time Car Part Instance Segmentation: the Comparison of the State-of-the-Art

YOLOv8, a real-time one-stage model, achieved high-precision car part segmentation with a mean Average Precision score of 0.70 and mAP-50 score of 0.89.

Automatyczne rozpoznawanie elementów polskich pojazdów hi storycznych

This study presents an automatic recognition system using artificial neural networks for Polish historical vehicles, achieving up to 100% classification accuracy in two cases.

Automatic recognition of dangerous objects in front of the vehicle’s windshield

This study proposes an automatic recognition system for dangerous objects in front of vehicle windshields using video recordings, with a best-case detection efficiency of $95 %.

Automated Vehicle Damage Detection and Repair Cost Estimation Using Deep Learning

The Vehicle Damage Assessment System (VDAS) accurately detects and estimates repair costs for vehicles using deep learning techniques, enabling informed decisions for insurance companies, motor workshops, and vehicle owners.

Accept the Modality Gap: An Exploration in the Hyperbolic Space

Accepting the modality gap between text and images in machine learning can preserve unimodal hierarchies and improve text-to-image and image-to-text retrieval tasks.

Structured Instance Understanding with Boundary Box Relationships in Object Detection System

Structured Instance Understanding (SIU) improves object detection systems by validating structural relationships within bounding boxes, increasing Mean Average Precision and F1-scores compared to traditional confidence score methods.

A System for Automated Vehicle Damage Localization and Severity Estimation Using Deep Learning

Our deep learning system accurately localizes and estimates vehicle damage in post-accident photos, reducing the time and effort needed for post-accident assessments.

CarAI: Car Inspection with Artificial Intelligence

CarAI revolutionizes car damage inspection processes by automating and enhancing them, leading to time-saving benefits, cost reductions, and minimizing subjective interpretations.