tensorflow-research

Goal

The goal of this repository is to research TensorFlow for machine vision industrial usage.
It will provide easy to use methods to train models from scratch and also fine tune them afterwards.

Starting point

To quickly get into the usage you can look into the notebooks, they provide the whole training pipline step by step.

Backbones

Backbones provide you with a ready to use model architectur.

Anomaly detection

Anomaly detection will use a autoencoder approche, the prediction substratced from the original image should show the anomaly.

Segmentation

Unet

• Vanilla Unet (original paper)
• Custom/Satellite Unet

Class	Group	Backbone	Sample notebook
ClassificationConv	Classification	classification_conv
	---	---
AutoEncoderConv	Anomaly Detection	auto_encoder_conv	Choco waffle anomaly detection sample
AutoEncoderFullConnected	Anomaly Detection	auto_encoder_full_connected
	---	---	---
SegmentationVanillaUnet	Segmentation	segmentation_vanilla_unet	Wood plate segmentation sample
TODO	Segmentation	satellite_unet
TODO	Segmentation	custom_unet