KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Settings

Author: search.filters.author.Kyu, Phyu Mar

Search Results

Now showing 1 - 2 of 2
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Advancing GAN Evaluation: The Advanced Mahalanobis Distance Learning Metric for Realistic Car Damage Image Assessment
    (2026-01-01)
    Kyu, Phyu Mar
    ;
    Woraratpanya, Kuntpong
    Generative Adversarial Networks (GANs) have demonstrated remarkable capability in synthesizing high-quality images from limited data, addressing challenges of data scarcity and diversity in deep learning (DL) training. This is particularly valuable for car damage classification, where real-world datasets are often limited. To mitigate this, we created a custom damaged-undamaged car dataset for training GAN models and generating realistic car damage images. However, evaluating the per-image realism of GAN-generated images remains challenging. Standard GAN metrics, such as Fréchet Inception Distance (FID), Kernel Inception Distance (KID), and Inception Score (IS), provide dataset-level scores but do not assess individual image quality. Meanwhile, Image Quality Assessment (IQA) metrics require reference images, rendering them unsuitable for reference-free scenarios, particularly in unpaired GAN-generated data. To address these limitations—including the practical failure of standard Mahalanobis Distance Learning (MDL) on small or high-dimensional datasets due to non-invertible covariance matrices—we propose Advanced Mahalanobis Distance Learning (AMDL), which incorporates adaptive regularization and pseudo-inverse refinement on deep feature embeddings from pre-trained CNNs. AMDL enables stable and reliable per-image realism assessment under covariance matrix instability, without requiring large datasets or ground-truth references. Our comprehensive evaluation framework involves three procedures: (1) dataset-level evaluation of four GAN models using standard GAN metrics, (2) per-image realism assessment with AMDL, and (3) classifier-based validation with CNN and Vision Transformer (ViT) models (with vs. without AMDL). Experimental results show that AMDL provides precise per-image realism assessment, outperforms existing GAN metrics across datasets, and offers a practical solution for evaluating unpaired GAN-generated images in car damage classification.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Car Damage Detection and Classification
    (2020-07-01)
    Kyu, Phyu Mar
    ;
    Woraratpanya, Kuntpong
    Nowadays, the proliferation of automobile industries is directly related to the increasing number of car incidents. So, insurance companies are facing many simultaneous claims and solving claims leakage. The sense of Artificial Intelligence (AI) based on machine learning and deep learning algorithms can help to solve these kinds of problem for insurance industries. In this paper, we apply deep learning-based algorithms, VGG16 and VGG19, for car damage detection and assessment in real-world datasets. The algorithms detect the damaged part of a car and assess its location and then its severity. Initially, we discover the effect of domain-specific pre-trained CNN models, which are trained on an ImageNet dataset, and followed by fine-tuning, because some of the categories can be fine-granular to get our specific tasks. Then we apply transfer learning in pre-trained VGG models and use some techniques to improve the accuracy of our system. We achieve the accuracy of 95.22% of VGG19 and 94.56% of VGG16 in the damaged detection, the accuracy of 76.48% of VGG19 and 74.39% of VGG16 in damage localization, the accuracy of 58.48% of VGG19 and 54.8% of VGG16 in damage severity with the combination of transfer learning and L2 regularization. From their results, the performance of VGG19 is better than VGG16. After analyzing and implementing our models, we find out that the results of using transfer learning and L2 regularization can work better than those of fine-tuning.