I am doing Task 1 from the challenge for Machine learning engineering position. Task is done using Python and Pytorch In this task,
- cut the images in vertically half.
- These vertically half images are shuffled and passed to convolutional autoencoder for training.
- After training, last layer from encoder network, Extracted the embeddings to get latent vector.
- This latent vector is used with K-nearest neighbor algorithm to find similar images.
- Similar images are found
- These similar images then restored with each other to find the original image before cutting it to half.
- The matching percentage is less.