• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    迪恩网络公众号

pankajmishra000/VT-ADL: A Vision Transformer Network for Image Anomaly Detection ...

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

pankajmishra000/VT-ADL

开源软件地址(OpenSource Url):

https://github.com/pankajmishra000/VT-ADL

开源编程语言(OpenSource Language):

Python 100.0%

开源软件介绍(OpenSource Introduction):

VT-ADL : A Vision Transformer Network for Image Anomaly Detection and Localization

Authors - Pankaj Mishra, Ricardo Verk, Daniele Fornasier, Claudio Piciarelli, Gian Luca Foresti

Abstract- We present a transformer-based image anomaly detection and localization network. Our proposed model is a combination of a reconstruction-based approach and patch embedding. The use of transformer networks helps preserving the spatial information of the embedded patches, which is later processed by a Gaussian mixture density network to localize the anomalous areas. In addition, we also publish BTAD, a real-world industrial anomaly dataset. Our results are compared with other state-of-the-art algorithms using publicly available datasets like MNIST and MVTec.

Network

The network is inspired from Vision Transformer. It adapts the trasnformer network for image anomaly detection and localization.

BeanTech Anomaly Detection Dataset - BTAD

Source: BeanTech srl License type: CC-BY-SA

Dataset contains RGB images of three industrial products – Scan to download

The images are captured from the industrial image acquisition systems and then cropped and log transformation was applied to respect privacy policy of the data owner(Beantech). Later pixel precise ground truth has been added by manually annotating the data using commercially available annotation tool "SuperAnnotate"

  • Product 1 : Contains 400 images of 1600x1600 pixels
  • Product 2 : Contains 1000 images of 600x600 pixels
  • Product 3 : Contains 399 images of 800x600 pixels

Results

  • MVTec Dataset - Real world anomaly dataset. contains 5354 high-resolution color and grey images of different texture and object categories.

  • BTAD Dataset - Consists of high resolution 1.8K RGB images of industrial products.

Ablation

  • Choice of number of Gaussian’s in the mixture model is justified with increasing number of Gaussian’s.
  • PRO Score first increases and then becomes constant

Regularization

  • Gaussian noise has been added to the encoded features from the transformer for regularization.
  • With Noise added the PRO score is 0.897 in contrary to 0.807 without noise.

Train (Command Line)

python train.py -p "hazelnut"

Cite

If you use this dataset, please cite it using the following reference:

P. Mishra, R. Verk, D. Fornasier, C. Piciarelli, G.L. Foresti
"VT-ADL: A Vision Transformer Network for Image Anomaly Detection and Localization"
30th IEEE/IES International Symposium on Industrial Electronics (ISIE)
Kyoto, Japan, June 20-23, 2021

BibTeX:


@inproceedings{

        mishra21-vt-adl,
        author = {Mishra, Pankaj and Verk, Riccardo and Fornasier, Daniele and Piciarelli, Claudio and Foresti, Gian Luca},
        title = {{VT-ADL}: A Vision Transformer Network for Image Anomaly Detection and Localization},
        booktitle = {30th IEEE/IES International Symposium on Industrial Electronics (ISIE)},
        year = {2021},
        month = {June},
        location = {Kyoto, Japan}
	}



鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap