Model Training Methods¶

What is it¶

When to use it¶

The AETrainer class should be used when you have lots of images, have very large images, or your given speed requirements are strict. Images can be thousands of pixels tall and wide, and can have multiple channels, which makes them very large. Encoding these images using an autoencoder can shrink the images significantly.

Theory behind it¶

AE Outlier detection¶

The encoder is trained to create dense embeddings for the images while the decoder is trained to reconstruct the new embedding into the original input image. The distances from the reconstructions between the test images and original images, or the probability distribution differences are used to measure how different they are and allow for the detection of Outliers.