Improve your segmentation

We are employing a data-centric approach (Andrew Ng, 2021) to help you develop a robust ML segmentation model as fast as possible. Our goal is to provide you with the necessary tools to efficiently create the ideal training dataset through an iterative process. This involves creating a dataset with just enough annotations in the most important locations to achieve the desired level of segmentation robustness.
In particular for complex tasks, start with a simple subset of your task and add complexity while you build your annotated dataset:

1. Start with one class that you want to segment
2. Annotate (at first around 50) objects/regions in images that seem similar (e.g. from one experiment)
3. Train and see if the segmentation of that class is sufficiently accurate
4. To make the algorithm more robust, add images with more variability (e.g. from different experiments) and repeat step 2 and 3
5. If the first class is segmented well across all images, iteratively annotate and train all other classes

If your objective is to develop a model that performs well across various imaging conditions, it's essential that your final annotated dataset reflects the variability that is expected in future data. Inter-dataset variability can arise due to various experimental setups or acquisition parameters, such as illumination, magnification, exposure, and differences in samples. Thus, to create a robust segmentation model, it's crucial to include examples that cover this variability in your annotated dataset.

Tips

• Don't waste your time annotating objects/regions that the algorithm has already learned to segment.
• Inspect the segmentations from your most recent training to learn which regions the algorithm couldn't segment and focus your annotations on these regions.
• Add a background border with at least one-pixel thickness between objects if the algorithm has trouble separating them.
• Rare classes are harder to learn. Try finding more training images to provide more examples of rare or other hard-to-learn cases.

Have questions? We're here to help.

Couldn't find what you are looking for? Reach out to our support team.