Interactive-Auto-Image-Labelling-with-Active-Learning

What does repository involve?

• Reading and pairing YOLO inputs as image objects-label pairs

• Denormalize YOLO cordinates for active leraning training

• Cropping each object in images, for my case it was tons of beers in freezer

• Resizing and reduction of dimensionality of images from rgb to grayscale

• Stacking resized image array from 200,50 to 10000 as MNIST dataset

• Normalize the array values for better Accuracy

This is an ongoing project. Stream-based pooling is used. After 800 query-epoch the accuracy is converged at =>
Current Accuracy: 0.4428 for 48 beer classes.

• Updates !! I added keras network with bayesian optimizer.
• Please find the labeler for nlp dataset. It uses logistic regression, include comprasion of entropy, uncertainity and random sampling with pool-based and stream-based apporaches.