lib-mapper-tf-model

Design and training the tflite model

Architecture

there will be two models

• segmentation: which shows each book-blob/cluster seperately in the image
• book recognition: which will finally map the book-spine image to its dataset

we will use siamese net to generate embbedings

tech-concerns

• if we use siamese embedding and keep including new books, the embeddings for all books may keep changing and huge database update may be requitred each time new books are included