It's been a while since deep learning has been introduced into our analysis protocols and we more and more use some pretty god models as Stardist or those from Cellpose2 (cyto, cyto2 nucl, etc.). But what if the model don’t work properly ? The logical next step will be to improve the model or to create new ones corresponding better to our images. In this workshop we will first work on the 2 most important part of the model creation/improvement by presenting in a first part some methods and some tools to do 2D annotations on different kind of 2D datas and how to use/export them. Then, in a second part, we’ll see how to evaluate the Model through It’s learning curve reading for the evaluation of the neural network behavior and through metrics for the prediction precision. For these purpose we’ll use QuPath and Cellpose2 and BIOP QuPath-cellpose2 plugin who are open source and provide very efficient annotations tools model creation and Quality Control metrics.