VIAE - Extending an Image Annotation Tool for Machine learning tasks

The talk is about building a web annotation tool for machine learning tasks such as image segmentation. I'll demonstrate how adapting a JS open source front-end, and building a Python server, interacting with S3, delivers a simple useable product. There will be a demo how the annotation tool can be used, and I'll comment on how I approached the development of the app.

Transcript

1. VIAE Extending an Image Annotation Tool for Machine Learning Tasks Louise Klodt Pyladies Dublin, 16 October 2018
2. The Project Software Engineering today: working with other people’s code. Goal: in-house image annotation tool for Machine learning tasks. To be used by company staff. Neural Networks are trained with artificial images. Use case: annotate a small set of real images for validation: Find out how Neural Network performs. Constraint: reduce development cost and time.
3. VIA Developed at University of Oxford (by Abhishek Dutta, Ankush Gupta, Andrew Zisserman). Open source. Some features of VIA: based solely on HTML, CSS and Javascript (no external javascript libraries). runs in the browser. import/export of region data in csv and json file format. image segmentation and classification.
4. Extending VIA VIA: 9000+ LOC Use a wrapper script Add a backend (Python) Use standard JSON format for image segmentation (COCO) Data transformations Storage: S3
5. Demo
6. Conclusion Never change a running system. You learn a lot by adapting someone else’s code.
7. Thank You!
8. References VGG Image Annotator (VIA): http://www.robots.ox.ac.uk/~vgg/software/via/ VIAE: https://github.com/LouiseKlodt/viae