With this web application, you and your team can collectively teach an image classifier
using individual interfaces. Each team member will have access to different features
organized into pages:
- Homepage - Here, you can view your model's performance as a team, including charts showing
the evolution of classes and accuracy rate.
- Navbar - This feature helps you track the model's training progress with a status bar,
and allows you to reload it or access settings.
- Teach - This section enables you to input images, view the model's predictions, and
add them to your dataset, with the option to correct associated labels.
- Share - Here, you can access the images you've personally added and assess the impact
of these additions on the model's training by recalculating its performance with the collective
dataset.
- Dataset - This section provides access to all instances contained in the collective
dataset, organized by label.
Additionally, you can access more advanced features to analyze the model's decisions on the
collective dataset:
- Inspector - Utilizing a confusion matrix, you can gain insight into how the collective
dataset is labeled by the model. By clicking on specific categories, you can further inspect
True and Predicted labels, along with confidence plots for selected images.
- Chat - Here, you have the embedded possibility to exchange information about your personal
data and send custom messages to debate or discuss aspects of the model's decisions or the
progression of your teaching as a group.