Edge Impulse Inc. / Visual crack detection Public

Target: Cortex-M7 216MHz Clone this project

Welcome to Edge Impulse, the largest community of edge AI developers!
This is a public Edge Impulse project, use the navigation bar to see all data and models in this project; or clone to retrain or deploy to any edge device.

Visual crack detection

Dataset acquired from https://digitalcommons.usu.edu/all_datasets/48/

Images

About this project

Visual Anomaly Detection (FOMO-AD)

Dataset: https://digitalcommons.usu.edu/all_datasets/48/

Maguire, M., Dorafshan, S., & Thomas, R. J. (2018). SDNET2018: A concrete crack image dataset for machine learning applications. Utah State University. https://doi.org/10.15142/T3TD19

Data (1,308 items)

View all

nominal.7106-16.jpg.3cpc87ql.ingestion-6c4b8994d8-dx5m8.jpg.42kn9g4k.d80bf3d42cc2

ID: 506297073

Label: nominal

nominal.7072-249.jpg.3cpc7q8d.ingestion-6c4b8994d8-d9rzn.jpg.42kn98r8.d80bf3d42cc2

ID: 506293620

Label: nominal

nominal.7135-106.jpg.3cpc8ggo.ingestion-6c4b8994d8-pvc99.jpg.42kn9m9q.d80bf3d42cc2

ID: 506300851

Label: nominal

nominal.7082-107.jpg.3cpc80dh.ingestion-6c4b8994d8-d9rzn.jpg.42kn9b4d.d80bf3d42cc2

ID: 506294443

Label: nominal

nominal.7138-85.jpg.3cpc8hb2.ingestion-6c4b8994d8-dx5m8.jpg.42kn9mte.d80bf3d42cc2

ID: 506301232

Label: nominal

nominal.7098-66.jpg.3cpc856i.ingestion-6c4b8994d8-pvc99.jpg.42kn9ea4.d80bf3d42cc2

ID: 506296239

Label: nominal

nominal.7097-75.jpg.3cpc84q5.ingestion-6c4b8994d8-pvc99.jpg.42kn9e0i.d80bf3d42cc2

ID: 506296086

Label: nominal

nominal.7074-119.jpg.3cpc7rgo.ingestion-6c4b8994d8-d9rzn.jpg.42kn995c.d80bf3d42cc2

ID: 506293721

Label: nominal

Model accuracy (Unoptimized (float32))

Validation set

Test set

95.9%

On-device performance (Cortex-M7 216MHz, Quantized (int8))

Latency

105 ms.

Peak RAM usage

625.6K

Flash usage

69.5K

Run this model

Scan QR code or launch in browser

Launch in browser

Dataset summary

Data collected

1,308 items

Labels

nominal

Project info

Project ID	288658
License	BSD 3-Clause Clear
No. of views	195,770
No. of clones	64