There are 3 different ways of applying .aiviadl models in Aivia:
As preprocessing step. You can use this as a step previous to apply a recipe. In the input/output region, it is possible to select a deep learning model to be applied before the recipe. Click on the folder icon, select your .aiviadl model, and the model will be applied before the recipe.
You can also see what the model does to the image without executing the recipe, by clicking on the magnifier icon for preview that will show up once the model and images are loaded.
Apply Deep Learning Recipe. By drag’n’drop into the viewer any of the models, you will enter in the Apply Deep Learning Recipe. The only purpose of this recipe is to apply the model to the image selected, by selecting the input channels. Check below for more information.
Apply Deep Learning for Aivia Models
In the Deep Learning Model Library https://www.aivia-software.com/model-library , you will find a collection of models ready to be used. These are pretrained models which means some microscopy data was collected to train those models with the goal of generalize a certain task (e.g., restore, label, segment,…).
You will find two types of models. First are Aivia Models, an extensive collection of models trained and developed by the Aivia Team. Second, the zoo models, developed externally, and at the moment, coming from bioimage model zoo (bioimage.io).
Load your image, drag the model, and click on start.
You can follow the video:
Good to know:
Aivia models can be used in workflows without problems.
Aivia models can be used with the Deep Learning Processor as pretrained.
Aivia models have been internally tested and validated.
These models have been imported from the bioimage.io webpage and adapted so that users can drag and drop the models directly into the Recipes window without needing to setup Python/CUDA environment.
At the moment we offer 4 models (bacterial segmentation, pancreatic cell segmentation, HPA cell segmentation, HPA nuclei segmentation) and we will continuously add additional models. If you want to know more information about the specific model, click on the question mark icon. You will find more information such as the author, DOI, related publications and licenses.
Inform yourself about the model (use the information panel), and find out if your data is appropriate for the model (scale, type, bit-depth, input channels). If the image is a 3D stack, formed by 2D images, and the model is 2D, use the Swap Z-T in the analysis options.
IMPORTANT!!!! Bioimage.io models cannot be used in workflows, as preprocessing or in the Deep Learning Processor.
Alternatively there is no problem if you save or keep the result of the model (in a channel) and use it later as input for any recipe.
Currently we are working to make more models available, and also to make them functional with other features present in Aivia (Deep Learning Processor, preprocessing etc).
Please be patient with the Bioimage.io execution, and write us if you experience problems or have any comments or wishes! We are happy to hear about it and it motivates us to keep improving!