DefectLens

A web platform where companies upload data to auto-generate a classification model that gives predictions and confidence scores for images. Includes an AI chatbot, and list of most common defects.

Comment

Inspiration

We want to support small local businesses in expanding their business by providing a website, where they can save costs by reducing human workload and raise efficiency for their business

What it does

A Website where companies can upload a file of a dataset, where a python automation creates a classification model that will be trained in under 3 minutes, and then can be used directly on the website by submitting images to the model and see the model's prediction and confidence score. The website will store data on different types of datasets that the user provides. (Attempting post hackathon implementation to be able to retrain the model by users "agreeing or disagreeing" with the model's prediction and telling the model which category it should go in to and also can be configured to a live camera footage on an assembly line instead of images). Website also provides the most common defect type along with an embedded chatbot, to research the defect, possible causes, and solutions.

How we built it

We started with creating the idea. We had a team of 3, one member worked on the frontend website. The other two worked on the backend configuration and classification model. Then worked together to integrate the backend and frontend.

Challenges we ran into

Attempting to connect the backend to the frontend created a challenge, since all the implementations would never match perfectly and code would be messed up or lost when trying to configure.

Accomplishments that we're proud of

We're proud that we were able to build a decently accurate classification model and add useful features in a simple to use website.

What we learned

We learned how to create a python automation to make a classification model and use public datasets to train the model. We also learned how to create a website and make it look appealing using HTML. We also learned how to connect the backend and frontend parts of the code properly.

What's next for Dataset Defect Detection Website

Attempting to implement a feature to be able to retrain the model by users "agreeing or disagreeing" with the model's prediction and telling the model which category it should go in to. Also can be configured to a live camera footage on an assembly line instead of images

Built With

Share this project:

Updates