BiaPy - accessible deep learning for bioimage analysis

Lecture¶

Daniel Franco-Barranco (MRC Laboratory of Molecular Biology, Cambridge)

Hands-on session¶

Google Colab¶

You will need a Google account to run this notebooks. Alternatively you can make use of Research Cloud.

Run 2D Instance Segmentation notebook¶

• Go to: https://biapyx.github.io/

• Select Colab notebooks

• Select 2D Instance Segmentation

• Open Google Colab, login with your Google account

• Install BiaPy (it might ask to restart session to install CUDA)

• At Manage File Sources pick Option 3: Download an Example Dataset

• Skip the OPTIONAL BioImage Model Zoo models check, it will take a long time to load

• Define parameters

• Start training

• Check the results of the training Try run the training with a low number of epochs first (e.g. 10 ) Check the Loss and IoU curves and visualize results

Run with a larger number of epochs, check if the results look better.

• Download model and configuration

Inference notebook¶

If you now like to use your model on other data you can use a different notebook:

• Upload some image (with or without GT)

• Model selection: Option 1

• Run inference

• Download the results

Run on SURF research cloud¶

Alternatively you can use biapy on Research Cloud. Access one on the desktops. Open JuyterLab

A biapy notebook that works on SRC can be found in this repository. Clone and select the biapy kernel.

Run biapy locally¶

conda create -n nlbi26-day3-biapy python=3.13
conda activate nlbi26-day3-biapy
pip install "biapy>=3.7.1" pyyaml gdown ipywidgets matplotlib

Then open BiaPy_instance_segmentation.ipynb in JupyterLab with this environment as kernel.