Photo Classification Evaluation Tool Students: Keren Bercu & Idan Cohen Supervisors : Tomer Shalev & Omer Geiger Industrial Project (234313)
Team Idan Cohen Keren Bercu
definitions Ground Truth (GT) - tagged sets of photos Dataset – A directory of images and a json file that contains the images path and set of tags Instance – An image
Goals Implement a full-stack web-based GUI tool The GUI will be used for: Creating dataset by uploading images and inserting the set of relevant tags Editing a dataset by labeling each instance Evaluating a classification Algorithm by comparing it to the corresponding GT Design a convenient friendly UI
Methodology The tool was developed with the intent of allowing Shutterfly engineers to later integrate to a full web environment (i.e. AWS: EC2, S3). Software development was based on short iterations with tight feedback loops (Agile concept). The project was managed on the GitHub platform, in order to coordinate, track, and update our work in one place (using issues and milestones throughout the way). The tool will be used to evaluate a variety of image classification algorithms.
achievements Web-based GUI with the following abilities: Dataset Uploader The Dataset Uploader allows the easy creation of a dataset with a set of tags and images Dataset Manager: The Dataset Manager allows to load and save a dataset with tagged instances Evaluator: The Evaluator compares between a GT set and the output of a classification algorithm. The Evaluator creates a confusion matrix as well as a display of the images along with the result of the comparison for each instance.
examples
Conclusions Integration between backend and frontend seems hard but in reality it is straight forward. Using Agile concept helped us achieve milestones in an efficient way. Full-stack development requires knowledge and an understanding of several disciplines.
Questions?
Getting familiar with: Week Ramp-Up: Getting familiar with: Python & Flask framework, Javascript, CSS, Vue.JS & Vuetify framework, JSON Setting up environment: Ubuntu VM, installations, GitHub 01/04 – 07/04 08/04 – 14/04 Build simple “Hello world”: Backend Frontend 15/04 – 21/04 22/04 – 28/04 Photo classification evaluation tool: Build a full dataset manager form Build a full functioning frontend 29/04 – 05/05 06/05 – 12/05 13/05 – 19/05 Load/Save integration: Triggering a connection between the frontend and the backend by clicking Load/Save 20/05 – 26/05 Upload integration: Create and implement an option to upload a folder of images and a schema file from the client to the server 27/05 – 02/06 Algorithm evaluator UI: Adding an evaluating functionality to the frontend 03/06 – 09/06 Integration & Implementation of algorithm evaluator UI 10/06 – 16/06 17/06 – 23/06 Bug fixes & documentation 24/06 – 30/06