Scientific output related to machine learning (ML) for imaging application has risen sharply since late 2017. Deep learning (a sub-set of ML) has been used in various applications from improving image resolution to de-noising and enhancing images to automatic image segmentation to predicting protein/organelle localization. W have developed a general framework for training and applying deep learning models using Aivia and Aivia Cloud. Leveraging the powerful cloud infrastructure by Google Cloud Platform, Aivia Cloud lets you train and apply deep learning models on your data. Sign up now to get the latest information about Aivia Cloud and for a spot in our Cloud beta program. No coding or AI expertise needed.
What is Aivia Cloud
Aivia Cloud (launched in parallel with Aivia 8) is a full-featured cloud-based image processing and visualization platform. It combines data storage, cloud computing, image processing and remote access. With Aivia Cloud, you can tap into virtually unlimited fast storage and state-of-the-art CPU and GPU hardware from anywhere with an internet connection.
Upload your data from your system to the cloud and access the data anywhere. Your stored data is safe and always available.
Aivia Cloud provisions high-power computing hardware with up to eight (8) NVIDIA V100 GPUs for training and applying deep learning models. A typical training run takes between 30 minutes and 1.5 hours.
Use your Aivia license anywhere in the world using your desktop, laptop or tablet. All the Aivia Cloud Web app needs is an internet connection and a web browser (no plug-ins needed). Simply log in and use Aivia on a high-performance cloud computer.
Start with an image
With Aivia Cloud, creating your own deep learning model starts with a pair of images - raw data and example data (aka ground truth). The raw data is the images captured using your standard protocol, while the example data varies by application:
For restoration, the best quality image that can be acquired with the imaging setup
For segmentation, a binary image with the objects of interests annotated
For prediction, the fluorescent image showing the localization of the subcellular feature of interest
Pick a model
With Aivia Cloud you can train 3 classes of deep learning models: restoration, segmentation and prediction. You can pick a model and train it completely with your own data or add your training data to the pre-trained models in Aivia Cloud using Transfer Learning.
You can also pick an existing, pre-trained model in Aivia Cloud to apply to the updated images.
After applying your deep learning model the processed images will be ready to download (if you processed them using Aivia Cloud). You can then review the processed images and do further analysis using the classic Aivia desktop app or the Aivia Cloud Web app.
Aivia Cloud provides pre-trained deep learning models for three (3) imaging applications. You can apply the existing models to your data or use them as starting points to augment the deep learning model with your own data. The three (3) pre-trained models are:
Enhance signal-to-noise ratio of images acquired with low laser power. Good for long-term imaging of live samples.
Transform original image into a probability map for features of interest. Good for segmenting EM and non-fluorescent images that are challenging for conventional methods.
Create prediction image for localization of subcellular features. Good for prototyping imaging experiments on label-sparse or label-free assays.