This  year I worked a lot with the new NLP revolutionary neural network named Transformer. I used it a lot in my job to build a Chatbot.

 One day, I read a new paper from Facebook research called End-to-End Object Detection with Transformers (DETR). 

I was very impressed about how powerful a Transformer layer can be. It is the first paper which shows a simple architecture to detect objects, which comes to compete with the other models much more complicated.

I made research on the  Net for a TensorFlow implementation of the DETR model but I did not find anything. So, I decided to implement my own TensorFlow DETR architecture in my deep tools package.

GitHub repository

Work steel in progress